train_20260729_152426

README.md

@@ -17,10 +17,10 @@ a 6-observation cadence (3 ON / 3 OFF, ABACAD) into an 8-dimensional latent, and
 Forest** classifies the cadence's concatenated latents as a technosignature candidate or not.
 
 This repository carries the released model weights at stable filenames, versioned via git
-tags: training tags match the pipeline run's save tag (e.g. `final_v3`), and release tags
-(`vX.Y.Z`) mark blessed weights.
+tags: training tags match the pipeline run's save tag (e.g. `train_20260101_120000`), and
+release tags (`vX.Y.Z`) mark blessed weights.
 
-**Training tag**: `test_v26`
+**Training tag**: `train_20260729_152426`
 
 ## Files
 
@@ -35,10 +35,10 @@ tags: training tags match the pipeline run's save tag (e.g. `final_v3`), and rel
 
 | Parameter | Value |
 |---|---|
-| Training rounds | `2` |
-| Epochs per round | `2` |
-| Beta-VAE samples per round | `200` |
-| Random Forest samples | `200` |
+| Training rounds | `10` |
+| Epochs per round | `100` |
+| Beta-VAE samples per round | `499200` |
+| Random Forest samples | `99840` |
 | Curriculum schedule | `exponential` |
 | SNR base | `10` |
 | Initial SNR range | `40` |
@@ -54,10 +54,10 @@ The complete configuration is in `config.json`.
 
 | Metric | Value |
 |---|---|
-| ROC AUC | 0.7812 |
-| Average precision | 0.7904 |
+| ROC AUC | 1.0000 |
+| Average precision | 1.0000 |
 | Classification threshold | 0.99 |
-| Validation samples | 40 |
+| Validation samples | 19968 |
 
 ## Library versions
 
@@ -71,11 +71,11 @@ The complete configuration is in `config.json`.
 
 ## Usage
 
-Aetherscan inference downloads these weights by default when no local artifact paths are
-given (pin a version with `--hf-revision`):
+Pin this training tag with `--hf-revision` to download exactly these weights (a bare no-artifact
+inference download resolves to the latest `vX.Y.Z` release tag instead, never a training tag):
 
 ```bash
-python -m aetherscan.main inference --hf-revision test_v26 --inference-files <catalog.csv>
+python -m aetherscan.main inference --hf-revision train_20260729_152426 --inference-files <catalog.csv>
 ```
 
 ## Links & citation

config.json

@@ -1,19 +1,17 @@
 {
-  "paths": {
-    "data_path": "/datax/scratch/zachy/data/aetherscan",
-    "model_path": "/datax/scratch/zachy/models/aetherscan",
-    "output_path": "/datax/scratch/zachy/outputs/aetherscan"
-  },
   "db": {
     "get_connection_timeout": 60.0,
     "stop_writer_timeout": 10.0,
     "write_interval": 5.0,
-    "write_buffer_max_size": 100,
+    "write_buffer_max_size": 5000,
     "write_retry_delay": 1.0,
-    "flush_timeout": 10.0
+    "flush_timeout": 10.0,
+    "bulk_chunk_rows": 50000,
+    "bulk_queue_max_items": 32,
+    "stop_drain_timeout": 600.0
   },
   "manager": {
-    "n_processes": 32,
+    "n_processes": 96,
     "chunks_per_worker": 4,
     "pool_terminate_timeout": 10.0
   },
@@ -21,7 +19,11 @@
     "get_gpu_timeout": 5.0,
     "stop_monitor_timeout": 10.0,
     "monitor_interval": 1.0,
-    "monitor_retry_delay": 1.0
+    "monitor_retry_delay": 1.0,
+    "annotate_stages": true,
+    "dashboard_enabled": true,
+    "dashboard_port": 8501,
+    "benchmark_report_enabled": true
   },
   "logger": {
     "console_level": "INFO",
@@ -44,20 +46,51 @@
       3
     ],
     "beta": 1.5,
-    "alpha": 10.0
+    "alpha": 10.0,
+    "mixed_precision": false,
+    "regularization_active": false
+  },
+  "reproducibility": {
+    "seed": 11,
+    "tf_deterministic_ops": true,
+    "derived_rf_seed": 961975133
   },
   "rf": {
     "n_estimators": 1000,
     "bootstrap": true,
     "max_features": "sqrt",
     "n_jobs": -1,
-    "seed": 11
+    "seed": null,
+    "latent_variant": "z_mean",
+    "active_dims": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7
+    ],
+    "z_aug_draws": 4,
+    "active_units_threshold": 0.01,
+    "selection_max_fpr": 0.01,
+    "selection_bootstrap_rounds": 500,
+    "max_ece": 0.05,
+    "calibration_min_isotonic": 1000,
+    "calibration_active": false,
+    "calibration_method": null,
+    "val_selection_fraction": 0.5,
+    "val_calibration_fraction": 0.25,
+    "screen_recall_tolerance": 0.0
   },
   "gpu": {
     "num_replicas": null,
     "per_gpu_memory_limit_mb": null,
     "nccl_num_packs": 2,
-    "use_async_allocator": true
+    "use_async_allocator": true,
+    "gpu_thread_mode": "gpu_private",
+    "gpu_thread_count": 2
   },
   "data": {
     "num_observations": 6,
@@ -69,40 +102,34 @@
     "num_target_backgrounds": 45000,
     "background_load_chunk_size": 15000,
     "max_chunks_per_file": 1,
-    "inference_background_load_chunk_size": 50000,
-    "train_files": [
-      "real_filtered_LARGE_HIP110750.npy",
-      "real_filtered_LARGE_HIP13402.npy",
-      "real_filtered_LARGE_HIP8497.npy"
-    ],
-    "test_files": [
-      "real_filtered_LARGE_test_HIP15638.npy"
-    ],
-    "inference_files": null
+    "inference_background_load_chunk_size": 50000
   },
   "training": {
-    "num_training_rounds": 2,
-    "epochs_per_round": 2,
-    "num_samples_beta_vae": 200,
-    "num_samples_rf": 200,
+    "num_training_rounds": 10,
+    "epochs_per_round": 100,
+    "posterior_collapse_kl_epsilon": 0.01,
+    "min_active_units_fraction": 0.5,
+    "posterior_collapse_patience": 5,
+    "num_samples_beta_vae": 499200,
+    "num_samples_rf": 99840,
     "train_val_split": 0.8,
-    "per_replica_batch_size": 4,
-    "effective_batch_size": 20,
-    "per_replica_val_batch_size": 4,
+    "per_replica_batch_size": 128,
+    "effective_batch_size": 7680,
+    "per_replica_val_batch_size": 64,
     "signal_injection_chunk_size": 50000,
-    "data_gen_task_size": 256,
+    "data_gen_task_size": 64,
+    "round_array_dtype": "float16",
     "round_data_dir": null,
     "overlap_data_generation": true,
-    "keep_round_data": false,
+    "keep_round_data": true,
     "plot_injection_subsampling_count": 100000,
     "plot_injection_outlier_percentile": 99.0,
-    "latent_viz_num_cadences_per_type": 5,
+    "latent_viz_num_cadences_per_type": 960,
     "latent_viz_step_interval": 10,
     "latent_viz_umap_fit_max_samples": 100000,
     "latent_viz_umap_n_neighbors": [
       5,
       15,
-      30,
       50
     ],
     "latent_viz_umap_min_dist": [
@@ -120,6 +147,7 @@
     "shap_top_k_features_dependence": 48,
     "rf_decision_boundary_grid_size": 150,
     "rf_decision_boundary_max_points": 5000,
+    "min_val_auc": 0.0,
     "snr_base": 10,
     "initial_snr_range": 40,
     "final_snr_range": 10,
@@ -132,15 +160,15 @@
     "min_pct_improvement": 0.001,
     "patience_threshold": 3,
     "reduction_factor": 0.2,
-    "max_retries": 1,
+    "max_retries": 3,
     "retry_delay": 60
   },
   "inference": {
-    "encoder_path": null,
-    "rf_path": null,
-    "config_path": null,
     "per_replica_batch_size": 2048,
     "classification_threshold": 0.99,
+    "screening_threshold": 0.5,
+    "mc_draws": 32,
+    "reference_cloud_size": 10000,
     "cadence_group_by_cols": [
       "Target",
       "Session",
@@ -151,7 +179,7 @@
     "cadence_h5_path_col": ".h5 path",
     "cadence_expected_obs": 6,
     "coarse_channel_width": 1048576,
-    "parallel_coarse_chans": null,
+    "coarse_channel_log_interval": null,
     "bandpass_method": "pfb",
     "pfb_taps_per_channel": 12,
     "bandpass_debug_plot": false,
@@ -165,7 +193,6 @@
     "overlap_fraction": 0.5,
     "discard_side_channels": false,
     "side_channel_count": 0,
-    "preprocess_output_dir": null,
     "inference_viz_enabled": true,
     "stamp_gallery_top_k": 12,
     "max_candidate_plots": 50,
@@ -178,10 +205,9 @@
     "revision": null
   },
   "checkpoint": {
-    "load_dir": null,
     "load_tag": null,
     "start_round": 1,
-    "save_tag": "test_v26",
+    "save_tag": "train_20260729_152426",
     "force_tag": false
   }
 }

random_forest.joblib

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13c7595b2d2ac27c851ae369cbc801069bfd481e52fbe2fbd46a80804e22b41d
-size 3882585
+oid sha256:91d01d786ed70678f3924991fb4e314bce7fd3ed4a5c44c47f670f526f43a81e
+size 15429465

vae_decoder.keras

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c9ea1fcaaff25fc0fde39c92057e51c6eb5f5e92410b5e7607aa56f4cbd11e2
+oid sha256:cb94a0a40183971b3415bd9f23c8b3792f850ec0bfd02426b9deac3682dedff4
 size 18677748

vae_encoder.keras

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e451ee8fbee81ef4608c6ab1e07d1c33916c726fb7e00f8489c9ab62fcad4d7
+oid sha256:9df2551c536cd70a86ff701f7a191a6d6bb3b650cb6b19c75acc9cdefc91c867
 size 18667989