src/split_data.py

import argparse
from pathlib import Path

import pandas as pd
from sklearn.model_selection import train_test_split

from utils import ensure_dir, save_json, set_seed


def main() -> None:
    parser = argparse.ArgumentParser(description="Create stratified train/val/test splits.")
    parser.add_argument("--input", required=True)
    parser.add_argument("--output_dir", required=True)
    parser.add_argument("--train_ratio", type=float, default=0.7)
    parser.add_argument("--val_ratio", type=float, default=0.1)
    parser.add_argument("--test_ratio", type=float, default=0.2)
    parser.add_argument("--seed", type=int, default=42)
    args = parser.parse_args()

    total = args.train_ratio + args.val_ratio + args.test_ratio
    if abs(total - 1.0) > 1e-8:
        raise ValueError("train_ratio + val_ratio + test_ratio must equal 1.0")

    set_seed(args.seed)
    data = pd.read_csv(args.input)
    ensure_dir(args.output_dir)

    train_val, test = train_test_split(
        data,
        test_size=args.test_ratio,
        random_state=args.seed,
        stratify=data["label_id"],
    )

    val_relative = args.val_ratio / (args.train_ratio + args.val_ratio)
    train, val = train_test_split(
        train_val,
        test_size=val_relative,
        random_state=args.seed,
        stratify=train_val["label_id"],
    )

    splits = {
        "train": train.sort_values("node_id").reset_index(drop=True),
        "val": val.sort_values("node_id").reset_index(drop=True),
        "test": test.sort_values("node_id").reset_index(drop=True),
    }

    for name, frame in splits.items():
        frame.to_csv(Path(args.output_dir) / f"{name}.csv", index=False)

    summary = {
        "seed": args.seed,
        "total_rows": int(len(data)),
        "ratios": {
            "train": args.train_ratio,
            "val": args.val_ratio,
            "test": args.test_ratio,
        },
        "splits": {
            name: {
                "rows": int(len(frame)),
                "label_distribution": frame["label"].value_counts().to_dict(),
            }
            for name, frame in splits.items()
        },
    }
    save_json(summary, Path(args.output_dir) / "split_summary.json")
    print("Saved train/val/test splits.")


if __name__ == "__main__":
    main()