scripts/build_classifier_dataset.py

"""Generate the SAT-classifier fine-tune dataset and write it to data/finetune/.

Run from the project root:  python -m scripts.build_classifier_dataset

Outputs:
  data/finetune/train.jsonl   — chat-format SFT examples
  data/finetune/val.jsonl
  data/finetune/stats.json    — coverage report

These files are ready to push to the Hub as a dataset (📡 Sharing is Caring).
"""

import sys
from collections import Counter
from pathlib import Path

try:
    sys.stdout.reconfigure(encoding="utf-8")
except (AttributeError, ValueError):
    pass

from src.finetune import build_examples, split, write_jsonl
from src.finetune.dataset import json


def main(per_phrasing: int = 8):
    out = Path("data/finetune")
    out.mkdir(parents=True, exist_ok=True)

    examples = build_examples(per_phrasing=per_phrasing)
    train, val = split(examples, val_fraction=0.1)

    write_jsonl(out / "train.jsonl", train)
    write_jsonl(out / "val.jsonl", val)

    by_account = Counter(e["label"]["cuenta"] for e in examples)
    by_kind = Counter(e["label"]["kind"] for e in examples)
    stats = {
        "total": len(examples),
        "train": len(train),
        "val": len(val),
        "accounts": len(by_account),
        "by_kind": dict(by_kind),
        "by_account": dict(by_account.most_common()),
    }
    (out / "stats.json").write_text(json.dumps(stats, ensure_ascii=False, indent=2),
                                    encoding="utf-8")

    print(f"Wrote {len(train)} train / {len(val)} val examples across "
          f"{len(by_account)} accounts → {out}/")
    print("By kind:", dict(by_kind))
    print("\nExample:")
    print(json.dumps(examples[0], ensure_ascii=False, indent=2))


if __name__ == "__main__":
    main()