scripts/validate_ani1x.py

#!/usr/bin/env python3
"""Validate the standardized ANI-1x dataset package."""

from __future__ import annotations

import argparse
import hashlib
import json
from pathlib import Path

import h5py


EXPECTED_SPLITS = {
    "ANI1x_cc_DFT_rc5_train": 8,
    "ANI1x_cc_DFT_rc5_val": 8,
    "ANI1x_cc_DFT_rc5_test": 8,
}
EXPECTED_FILES = {
    "ANI1x_cc_DFT_rc5_statistics.json",
    "ani1x_cc_dft.xyz",
    "ani1x_train.xyz",
    "ani1x_test.xyz",
}
REQUIRED_SAMPLE_KEYS = (
    "atomic_numbers",
    "cell",
    "pbc",
    "positions",
    "properties/energy",
    "properties/forces",
)


def sha256_file(path: Path) -> str:
    h = hashlib.sha256()
    with path.open("rb") as f:
        for chunk in iter(lambda: f.read(1024 * 1024), b""):
            h.update(chunk)
    return h.hexdigest()


def validate_h5(path: Path) -> None:
    with h5py.File(path, "r") as f:
        batches = sorted(f.keys())
        if not batches:
            raise SystemExit(f"[FAIL] HDF5 无 batch: {path}")
        configs = sorted(f[batches[0]].keys())
        if not configs:
            raise SystemExit(f"[FAIL] HDF5 batch 无 config: {path}")
        sample = f[batches[0]][configs[0]]
        for key in REQUIRED_SAMPLE_KEYS:
            if key not in sample:
                raise SystemExit(f"[FAIL] HDF5 缺少字段 {key}: {path}")
        atomic_numbers = sample["atomic_numbers"]
        positions = sample["positions"]
        forces = sample["properties/forces"]
        if atomic_numbers.dtype.kind not in ("i", "u"):
            raise SystemExit(f"[FAIL] atomic_numbers dtype 非整数: {path}")
        if positions.dtype.kind != "f" or positions.shape[-1] != 3:
            raise SystemExit(f"[FAIL] positions dtype/shape 异常: {path}")
        if forces.dtype.kind != "f" or forces.shape != positions.shape:
            raise SystemExit(f"[FAIL] forces dtype/shape 异常: {path}")


def validate_checksum_manifest(dataset_root: Path, manifest_path: Path) -> None:
    if not manifest_path.exists():
        raise SystemExit(f"[FAIL] 校验清单不存在: {manifest_path}")
    checked = 0
    with manifest_path.open("r", encoding="utf-8") as f:
        for line in f:
            line = line.strip()
            if not line or line.startswith("#"):
                continue
            digest, rel = line.split(maxsplit=1)
            path = dataset_root / rel
            if not path.exists():
                raise SystemExit(f"[FAIL] 清单文件不存在: {path}")
            actual = sha256_file(path)
            if actual != digest:
                raise SystemExit(f"[FAIL] SHA256 不一致: {rel}")
            checked += 1
    if checked == 0:
        raise SystemExit("[FAIL] 校验清单为空")
    print(f"[OK] SHA256 清单验证通过: {checked} 个文件")


def main() -> int:
    parser = argparse.ArgumentParser()
    parser.add_argument("--dataset-root", default="data/ani1x")
    parser.add_argument("--checksum-manifest", default="metadata/sha256_manifest.txt")
    args = parser.parse_args()

    repo_root = Path.cwd()
    dataset_root = Path(args.dataset_root)
    if not dataset_root.is_absolute():
        dataset_root = repo_root / dataset_root
    checksum_manifest = Path(args.checksum_manifest)
    if not checksum_manifest.is_absolute():
        checksum_manifest = repo_root / checksum_manifest

    if not dataset_root.exists():
        raise SystemExit(f"[FAIL] 数据目录不存在: {dataset_root}")

    for rel in EXPECTED_FILES:
        path = dataset_root / rel
        if not path.exists():
            raise SystemExit(f"[FAIL] 缺少文件: {path}")
        if path.stat().st_size <= 0:
            raise SystemExit(f"[FAIL] 文件为空: {path}")
        print(f"[OK] 文件存在: {rel}")

    stats_path = dataset_root / "ANI1x_cc_DFT_rc5_statistics.json"
    with stats_path.open("r", encoding="utf-8") as f:
        stats = json.load(f)
    for key in ("atomic_energies", "avg_num_neighbors", "mean", "std", "atomic_numbers", "r_max"):
        if key not in stats:
            raise SystemExit(f"[FAIL] statistics 缺少字段: {key}")
    print("[OK] statistics JSON 可读")

    for split, expected_count in EXPECTED_SPLITS.items():
        split_dir = dataset_root / split
        h5_files = sorted(split_dir.glob("*.h5"))
        if len(h5_files) != expected_count:
            raise SystemExit(f"[FAIL] {split} 分片数异常: {len(h5_files)} != {expected_count}")
        validate_h5(h5_files[0])
        print(f"[OK] {split}: {len(h5_files)} 个 h5 分片，样本结构可读")

    validate_checksum_manifest(dataset_root, checksum_manifest)
    print("[OK] ANI-1x 数据集读取验证通过")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())