Datasets:
File size: 1,959 Bytes
65b8e03 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 | #!/usr/bin/env python3
"""Sample N MathVision test items with a fixed seed and write images + items.jsonl.
Items without an image are excluded before sampling. Every item keeps its
official id, question, options, answer, level, subject, and image sha256.
"""
from __future__ import annotations
import argparse
import hashlib
import json
import random
from pathlib import Path
import pyarrow.parquet as pq
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--parquet", type=Path, required=True)
parser.add_argument("--out", type=Path, required=True)
parser.add_argument("--n", type=int, default=300)
parser.add_argument("--seed", type=int, default=20261106)
args = parser.parse_args()
rows = [r for r in pq.read_table(args.parquet).to_pylist() if r["decoded_image"] and r["decoded_image"]["bytes"]]
rows.sort(key=lambda r: int(r["id"]))
sample = sorted(random.Random(args.seed).sample(rows, args.n), key=lambda r: int(r["id"]))
image_dir = args.out / "images"
image_dir.mkdir(parents=True, exist_ok=True)
with (args.out / "items.jsonl").open("w", encoding="utf-8") as handle:
for r in sample:
data = r["decoded_image"]["bytes"]
suffix = Path(r["image"]).suffix or ".png"
path = image_dir / f"{r['id']}{suffix}"
path.write_bytes(data)
handle.write(json.dumps({
"item_id": f"mathvision_{r['id']}", "id": r["id"], "question": r["question"],
"options": r["options"], "answer": r["answer"], "level": r["level"], "subject": r["subject"],
"image": f"images/{path.name}", "image_sha256": hashlib.sha256(data).hexdigest(),
}, ensure_ascii=False) + "\n")
print(json.dumps({"pool": len(rows), "sampled": len(sample), "seed": args.seed,
"multiple_choice": sum(bool(r["options"]) for r in sample)}))
if __name__ == "__main__":
main()
|