Delete survhte_bench.py

survhte_bench.py

@@ -1,102 +0,0 @@
-"""
-HuggingFace dataset loading script for SurvHTE-Bench.
-
-Usage
------
-# Full data (default) — reconstruct experiment_setups / experiment_repeat_setups
-from datasets import load_dataset
-ds = load_dataset("snoroozi/SurvHTE-Bench", "synthetic")
-
-# Pre-computed splits — get a specific train/val/test fold
-ds = load_dataset("snoroozi/SurvHTE-Bench", "synthetic_splits")
-train0 = ds["train_0"].to_pandas()
-
-# Or use the helper wrappers (hf_load.py):
-from hf_load import load_data, load_splits
-experiment_setups, experiment_repeat_setups = load_data("synthetic")
-results = load_splits("synthetic")
-"""
-
-import datasets
-
-_DESCRIPTION = """\
-SurvHTE-Bench: A Benchmark for Heterogeneous Treatment Effect Estimation in Survival Analysis.
-Includes synthetic, semi-synthetic, and real-world datasets with pre-computed train/val/test splits.
-Published at ICLR 2026.
-"""
-
-_HOMEPAGE   = "https://huggingface.co/datasets/snoroozi/SurvHTE-Benchmark"
-_LICENSE    = "MIT"
-_REPO_URL   = "https://huggingface.co/datasets/snoroozi/SurvHTE-Benchmark/resolve/main"
-
-# ── dataset names ──────────────────────────────────────────────────────────
-_DATASETS = {
-    "synthetic": "synthetic",
-    "actg_syn":  "semi-synthetic",
-    "twin":      "real",
-    "actgHC":    "real",
-    "actgLC":    "real",
-}
-
-_ALL_CONFIGS = list(_DATASETS.keys()) + [f"{n}_repeats" for n in _DATASETS] + \
-               [f"{n}_splits"  for n in _DATASETS]
-
-
-class SurvHTEBenchConfig(datasets.BuilderConfig):
-    def __init__(self, name, **kwargs):
-        super().__init__(name=name, version=datasets.Version("1.0.0"), **kwargs)
-
-
-class SurvHTEBench(datasets.GeneratorBasedBuilder):
-    """SurvHTE-Bench dataset loader."""
-
-    BUILDER_CONFIGS = [SurvHTEBenchConfig(name=n) for n in _ALL_CONFIGS]
-    DEFAULT_CONFIG_NAME = "synthetic"
-
-    def _info(self):
-        # Features are discovered at runtime from parquet; return minimal info here.
-        return datasets.DatasetInfo(
-            description=_DESCRIPTION,
-            homepage=_HOMEPAGE,
-            license=_LICENSE,
-        )
-
-    def _split_generators(self, dl_manager):
-        name = self.config.name
-
-        # ── repeats config ─────────────────────────────────────────────────
-        if name.endswith("_repeats"):
-            parquet_url = f"{_REPO_URL}/{name}/train/0000.parquet"
-            path = dl_manager.download(parquet_url)
-            return [datasets.SplitGenerator(name=datasets.Split.TRAIN,
-                                            gen_kwargs={"path": path})]
-
-        # ── splits config ─────────────────────────────────────────────────
-        if name.endswith("_splits"):
-            base = name[: -len("_splits")]
-            splits = []
-            for split_type in ("train", "val", "test"):
-                for r in range(10):
-                    split_name = f"{split_type}_{r}"
-                    url  = f"{_REPO_URL}/{base}_splits/{split_name}/0000.parquet"
-                    path = dl_manager.download(url)
-                    splits.append(
-                        datasets.SplitGenerator(
-                            name=split_name,
-                            gen_kwargs={"path": path},
-                        )
-                    )
-            return splits
-
-        # ── full data config ───────────────────────────────────────────────
-        parquet_url = f"{_REPO_URL}/{name}/train/0000.parquet"
-        path = dl_manager.download(parquet_url)
-        return [datasets.SplitGenerator(name=datasets.Split.TRAIN,
-                                        gen_kwargs={"path": path})]
-
-    def _generate_examples(self, path):
-        import pandas as pd
-        df = pd.read_parquet(path)
-        for idx, row in df.iterrows():
-            yield idx, row.to_dict()
-