scripts/build_streamlit_aligned_parquet.py

#!/usr/bin/env python3
from __future__ import annotations

import argparse
import ast
import json
from collections import Counter
from pathlib import Path
import sys
from typing import Any

import pyarrow as pa
import pyarrow.parquet as pq

ROOT = Path(__file__).resolve().parents[1]
if str(ROOT) not in sys.path:
    sys.path.insert(0, str(ROOT))

import dataset_catalog as catalog


def _to_rel(base_dir: Path, value: str) -> str:
    text = str(value or "").strip()
    if not text:
        return ""
    p = Path(text)
    if not p.is_absolute():
        return p.as_posix()
    try:
        return p.resolve().relative_to(base_dir).as_posix()
    except ValueError:
        return p.as_posix()


def _build_rows(base_dir: Path, dataset: list[dict[str, Any]]) -> list[dict[str, Any]]:
    rows: list[dict[str, Any]] = []
    for item in dataset:
        row = dict(item)
        row["video_rel_path"] = str(item.get("path", "")).strip()
        row["image_gt_rel_path"] = _to_rel(base_dir, str(item.get("image_gt", "")).strip())
        image_paths = item.get("image_paths") if isinstance(item.get("image_paths"), list) else []
        row["image_paths_rel"] = [_to_rel(base_dir, str(path)) for path in image_paths]
        rows.append(row)
    rows.sort(key=lambda r: str(r.get("path", "")))
    return rows


def _load_default_descriptions(base_dir: Path) -> dict[str, str]:
    app_path = base_dir / "app.py"
    if not app_path.exists():
        return {}
    tree = ast.parse(app_path.read_text(encoding="utf-8"))
    for node in tree.body:
        if isinstance(node, ast.Assign):
            for target in node.targets:
                if isinstance(target, ast.Name) and target.id == "DEFAULT_DESCRIPTIONS":
                    value = ast.literal_eval(node.value)
                    if isinstance(value, dict):
                        return {str(k): str(v) for k, v in value.items()}
    return {}


def _validate_rows(base_dir: Path, rows: list[dict[str, Any]]) -> dict[str, Any]:
    missing_videos: list[str] = []
    for row in rows:
        rel = str(row.get("path", "")).strip()
        if not rel:
            continue
        if not (base_dir / rel).exists():
            missing_videos.append(rel)

    split_counter = Counter(str(row.get("data_split", "")).strip().lower() for row in rows)
    category_counter = Counter(str(row.get("category", "")).strip() for row in rows)
    major_counter = Counter(str(row.get("major", "")).strip() for row in rows)

    return {
        "total_rows": len(rows),
        "split_counts": dict(sorted(split_counter.items())),
        "category_counts": dict(sorted(category_counter.items())),
        "major_counts": dict(sorted(major_counter.items())),
        "missing_video_paths": missing_videos,
    }


def main() -> int:
    parser = argparse.ArgumentParser(
        description="Build Parquet aligned with Streamlit app _build_dataset(meta) output."
    )
    parser.add_argument(
        "--base-dir",
        default=str(Path(__file__).resolve().parents[1]),
        help="Repository base directory (default: script parent parent).",
    )
    parser.add_argument(
        "--out-parquet",
        default="dist/streamlit_aligned.parquet",
        help="Parquet output path relative to base dir.",
    )
    parser.add_argument(
        "--out-stats",
        default="dist/streamlit_aligned_stats.json",
        help="Stats output path relative to base dir.",
    )
    parser.add_argument(
        "--strict-files",
        action="store_true",
        help="Fail if any video file in rows is missing on disk.",
    )
    args = parser.parse_args()

    base_dir = Path(args.base_dir).resolve()
    video_dir = base_dir / "videos"
    dataset_test_path = base_dir / "video_dataset.json"
    dataset_train_path = base_dir / "video_dataset_train.json"
    legacy_path = base_dir / "video_texts.json"

    out_parquet = Path(args.out_parquet)
    if not out_parquet.is_absolute():
        out_parquet = (base_dir / out_parquet).resolve()
    out_stats = Path(args.out_stats)
    if not out_stats.is_absolute():
        out_stats = (base_dir / out_stats).resolve()
    out_parquet.parent.mkdir(parents=True, exist_ok=True)
    out_stats.parent.mkdir(parents=True, exist_ok=True)

    meta, _, _ = catalog.load_meta(
        base_dir=base_dir,
        dataset_test_path=dataset_test_path,
        dataset_train_path=dataset_train_path,
        legacy_path=legacy_path,
    )
    default_descriptions = _load_default_descriptions(base_dir)
    dataset = catalog.build_dataset(
        base_dir=base_dir,
        video_dir=video_dir,
        meta=meta,
        default_descriptions=default_descriptions,
    )
    rows = _build_rows(base_dir, dataset)
    stats = _validate_rows(base_dir, rows)

    if args.strict_files and stats["missing_video_paths"]:
        sample = stats["missing_video_paths"][:20]
        raise SystemExit(
            f"Missing {len(stats['missing_video_paths'])} video paths. Sample: {sample}"
        )

    table = pa.Table.from_pylist(rows)
    pq.write_table(table, out_parquet, compression="zstd")
    out_stats.write_text(json.dumps(stats, ensure_ascii=False, indent=2), encoding="utf-8")

    print(f"Rows: {stats['total_rows']}")
    print(f"Split counts: {stats['split_counts']}")
    print(f"Parquet: {out_parquet}")
    print(f"Stats: {out_stats}")
    if stats["missing_video_paths"]:
        print(f"Missing video paths: {len(stats['missing_video_paths'])}")
    return 0


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