scripts/tabdiff_prepare.py

"""Prepare a MemisisLabs dataset for TabDiff (github.com/MinkaiXu/TabDiff).

Writes the CSV + Info JSON into a TabDiff checkout's data/ layout so you can run
`process_dataset.py`, then train/sample. Column-type indices (num/cat/target) are
auto-detected from our dataset registry.

Usage (on the lab server, inside the TabDiff repo's conda env):
    python scripts/tabdiff_prepare.py --dataset openml_45040 \
        --name schizophrenia --tabdiff-root /path/to/TabDiff
"""
from __future__ import annotations

import argparse
import json
import sys
from pathlib import Path

sys.path.insert(0, str(Path(__file__).resolve().parent.parent))

from pipeline import datasets  # noqa: E402
from pipeline.metadata import is_categorical  # noqa: E402


def main() -> None:
    ap = argparse.ArgumentParser(description=__doc__)
    ap.add_argument("--dataset", default="openml_45040", help="MemisisLabs dataset id")
    ap.add_argument("--name", default="schizophrenia", help="TabDiff dataset name")
    ap.add_argument("--tabdiff-root", required=True, help="path to the cloned TabDiff repo")
    args = ap.parse_args()

    real, target, _protected, task = datasets.load(args.dataset)
    cols = list(real.columns)
    target_idx = cols.index(target)

    num_idx, cat_idx = [], []
    for i, c in enumerate(cols):
        if i == target_idx:
            continue
        (cat_idx if is_categorical(real[c]) else num_idx).append(i)

    root = Path(args.tabdiff_root)
    (root / "data" / args.name).mkdir(parents=True, exist_ok=True)
    (root / "data" / "Info").mkdir(parents=True, exist_ok=True)

    csv_path = root / "data" / args.name / f"{args.name}.csv"
    real.to_csv(csv_path, index=False)

    info = {
        "name": args.name,
        "task_type": "binclass" if task == "binclass" else "regression",
        "header": "infer",
        "column_names": None,
        "num_col_idx": num_idx,
        "cat_col_idx": cat_idx,
        "target_col_idx": [target_idx],
        "file_type": "csv",
        "data_path": f"data/{args.name}/{args.name}.csv",
        "test_path": None,
    }
    info_path = root / "data" / "Info" / f"{args.name}.json"
    info_path.write_text(json.dumps(info, indent=4))

    print(f"wrote {csv_path}  ({real.shape})")
    print(f"wrote {info_path}")
    print(f"  num_col_idx={num_idx}\n  cat_col_idx={cat_idx}\n  target_col_idx=[{target_idx}] ({target})")
    print("\nNext (in the TabDiff repo):")
    print(f"  python process_dataset.py --dataname {args.name}")
    print(f"  python main.py --dataname {args.name} --mode train --no_wandb")
    print(f"  python main.py --dataname {args.name} --mode test --report --no_wandb")


if __name__ == "__main__":
    main()