data/unsupervised_train_edge/explore_unsupervised_train.py

#!/usr/bin/env python3
"""Explore and extract IDAMA unsupervised_train sharded tar archives.

Data layout (this directory):
  patent/shard_XXXX.tar   product/shard_XXXX.tar
  manifest.jsonl          summary.json
  _filelists/<cat>/shard_XXXX.txt

Images are UAED edge maps (grayscale JPEG) from:
  /data2/gonghaifan/UAED/results/train

Examples:
  python explore_unsupervised_train.py summary
  python explore_unsupervised_train.py list-tar patent 0 --limit 10
  python explore_unsupervised_train.py sample --category patent -n 5 -o /tmp/preview
  python explore_unsupervised_train.py extract --category product --shard 0 -o /tmp/shard0
  python explore_unsupervised_train.py verify --category patent --shard 0
"""

from __future__ import annotations

import argparse
import json
import random
import sys
import tarfile
from collections import Counter
from pathlib import Path
from typing import Iterator

SCRIPT_DIR = Path(__file__).resolve().parent
DEFAULT_ROOT = SCRIPT_DIR


def load_summary(root: Path) -> dict:
    path = root / "summary.json"
    if not path.is_file():
        raise FileNotFoundError(f"Missing {path}")
    return json.loads(path.read_text(encoding="utf-8"))


def tar_path(root: Path, category: str, shard_id: int, gzip: bool = False) -> Path:
    ext = ".tar.gz" if gzip else ".tar"
    return root / category / f"shard_{shard_id:04d}{ext}"


def iter_manifest(
    root: Path,
    *,
    category: str | None = None,
    shard_id: int | None = None,
) -> Iterator[dict]:
    path = root / "manifest.jsonl"
    if not path.is_file():
        raise FileNotFoundError(f"Missing {path}")
    with path.open(encoding="utf-8") as fh:
        for line in fh:
            line = line.strip()
            if not line:
                continue
            rec = json.loads(line)
            if category is not None and rec.get("category") != category:
                continue
            if shard_id is not None and rec.get("shard_id") != shard_id:
                continue
            yield rec


def cmd_summary(args: argparse.Namespace) -> int:
    root: Path = args.root
    summary = load_summary(root)
    counts = summary.get("counts", {})
    plan = summary.get("shard_plan", {})
    print("=== unsupervised_train 数据集概览 ===\n")
    print(f"数据根目录:     {root.resolve()}")
    print(f"原始来源:       {summary.get('source', '?')}")
    print(f"打包输出:       {summary.get('output', '?')}")
    print(f"压缩格式:       {'tar.gz' if summary.get('gzip') else 'tar (未 gzip)'}")
    print(f"分片总数:       {summary.get('num_shards_requested', '?')}")
    print(f"全部校验通过:   {summary.get('all_shards_verified', '?')}")
    print()
    print("类别统计:")
    total = 0
    for cat in ("patent", "product", "other"):
        n = counts.get(cat, 0)
        if n:
            print(f"  {cat:8s}  {n:>10,} 张  ->  {plan.get(cat, 0)} 个 tar 分片")
            total += n
    print(f"  {'合计':8s}  {total:>10,} 张")
    print()
    print("命名规则:")
    print("  patent  US/CN/EP/... 专利号 + _序号.jpg，如 US10000225_1.jpg")
    print("  product product_img_... 产品边缘图")
    print()
    print("单条 manifest 字段: category, shard_id, tar, member, source")
    print()
    shards = summary.get("shards", [])
    if shards:
        print("分片体积 (前 3 / 后 3):")
        for rec in shards[:3] + shards[-3:]:
            mb = rec.get("tar_bytes", 0) / (1024 * 1024)
            print(
                f"  {rec['tar']:40s}  {rec['num_files']:>7,} 文件  {mb:>7.1f} MiB"
            )
    manifest = root / "manifest.jsonl"
    if manifest.is_file():
        print(f"\nmanifest.jsonl: {manifest.stat().st_size / (1024**2):.1f} MiB")
    return 0


def cmd_list_shards(args: argparse.Namespace) -> int:
    summary = load_summary(args.root)
    for rec in summary.get("shards", []):
        if args.category and rec["category"] != args.category:
            continue
        mb = rec.get("tar_bytes", 0) / (1024 * 1024)
        print(f"{rec['category']:7s}  shard {rec['shard_id']:4d}  "
              f"{rec['num_files']:7,} files  {mb:8.1f} MiB  {rec['tar']}")
    return 0


def cmd_list_tar(args: argparse.Namespace) -> int:
    root: Path = args.root
    summary = load_summary(root)
    gzip = bool(summary.get("gzip"))
    path = tar_path(root, args.category, args.shard, gzip=gzip)
    if not path.is_file():
        print(f"Not found: {path}", file=sys.stderr)
        return 1
    limit = args.limit
    with tarfile.open(path, "r:gz" if gzip else "r:") as tf:
        names = tf.getnames()
        for i, name in enumerate(names):
            if i >= limit:
                print(f"... ({len(names) - limit} more members)")
                break
            print(name)
    return 0


def cmd_extract(args: argparse.Namespace) -> int:
    root: Path = args.root
    summary = load_summary(root)
    gzip = bool(summary.get("gzip"))
    out: Path = args.output
    out.mkdir(parents=True, exist_ok=True)

    shards: list[tuple[str, int]] = []
    if args.category is not None and args.shard is not None:
        shards = [(args.category, args.shard)]
    elif args.category is not None:
        n = summary.get("shard_plan", {}).get(args.category, 0)
        shards = [(args.category, i) for i in range(n)]
    else:
        for rec in summary.get("shards", []):
            shards.append((rec["category"], rec["shard_id"]))

    members_filter: set[str] | None = None
    if args.members_file:
        members_filter = {
            line.strip()
            for line in Path(args.members_file).read_text(encoding="utf-8").splitlines()
            if line.strip()
        }

    extracted = 0
    for category, shard_id in shards:
        tar_p = tar_path(root, category, shard_id, gzip=gzip)
        if not tar_p.is_file():
            print(f"Skip missing: {tar_p}", file=sys.stderr)
            continue
        shard_out = out if args.flat else out / category
        shard_out.mkdir(parents=True, exist_ok=True)
        print(f"Extracting {tar_p} -> {shard_out}")
        with tarfile.open(tar_p, "r:gz" if gzip else "r:") as tf:
            for member in tf.getmembers():
                if not member.isfile():
                    continue
                if members_filter is not None and member.name not in members_filter:
                    continue
                dest = shard_out / Path(member.name).name
                if dest.exists() and not args.overwrite:
                    continue
                tf.extract(member, path=shard_out)
                extracted += 1
                if args.max_files and extracted >= args.max_files:
                    print(f"Reached --max-files {args.max_files}")
                    return 0
    print(f"Done. Extracted {extracted} file(s) under {out.resolve()}")
    return 0


def cmd_sample(args: argparse.Namespace) -> int:
    root: Path = args.root
    records: list[dict] = []
    # Reservoir-style: collect matching lines (may be large for full category)
    for rec in iter_manifest(root, category=args.category, shard_id=args.shard):
        records.append(rec)
        if args.pool_size and len(records) >= args.pool_size:
            break
    if not records:
        print("No manifest records matched.", file=sys.stderr)
        return 1

    if args.seed is not None:
        random.seed(args.seed)
    n = min(args.num, len(records))
    picks = random.sample(records, n) if len(records) > n else records

    out: Path = args.output
    out.mkdir(parents=True, exist_ok=True)
    summary = load_summary(root)
    gzip = bool(summary.get("gzip"))

    meta_path = out / "samples.jsonl"
    with meta_path.open("w", encoding="utf-8") as meta_fh:
        for rec in picks:
            tar_p = root / rec["tar"]
            member = rec["member"]
            cat = rec["category"]
            dest_dir = out / cat if not args.flat else out
            dest_dir.mkdir(parents=True, exist_ok=True)
            dest = dest_dir / Path(member).name
            with tarfile.open(tar_p, "r:gz" if gzip else "r:") as tf:
                fobj = tf.extractfile(member)
                if fobj is None:
                    print(f"Missing member {member} in {tar_p}", file=sys.stderr)
                    continue
                dest.write_bytes(fobj.read())
            meta_fh.write(json.dumps(rec, ensure_ascii=False) + "\n")
            info = inspect_image(dest)
            print(f"  {cat}/{member}  ->  {dest.name}  {info}")

    print(f"\nWrote {len(picks)} image(s) and {meta_path}")
    return 0


def inspect_image(path: Path) -> str:
    try:
        from PIL import Image
    except ImportError:
        return f"({path.stat().st_size} bytes, install Pillow for size/mode)"
    with Image.open(path) as img:
        return f"{img.size[0]}x{img.size[1]} mode={img.mode}"


def cmd_verify(args: argparse.Namespace) -> int:
    root: Path = args.root
    summary = load_summary(root)
    gzip = bool(summary.get("gzip"))
    shards = summary.get("shards", [])
    if args.category:
        shards = [s for s in shards if s["category"] == args.category]
    if args.shard is not None:
        shards = [s for s in shards if s["shard_id"] == args.shard]

    ok = True
    for rec in shards:
        cat, sid = rec["category"], rec["shard_id"]
        tar_p = tar_path(root, cat, sid, gzip=gzip)
        expected = rec["num_files"]
        if not tar_p.is_file():
            print(f"FAIL  missing tar: {tar_p}")
            ok = False
            continue
        with tarfile.open(tar_p, "r:gz" if gzip else "r:") as tf:
            n = sum(1 for m in tf.getmembers() if m.isfile())
        status = "OK" if n == expected else "MISMATCH"
        if n != expected:
            ok = False
        print(f"{status:8s}  {rec['tar']:35s}  tar={n:,}  expected={expected:,}")

        if args.check_manifest:
            m_count = sum(1 for _ in iter_manifest(root, category=cat, shard_id=sid))
            if m_count != expected:
                print(f"         manifest lines={m_count:,} (expected {expected:,})")
                ok = False
    return 0 if ok else 1


def cmd_manifest_lookup(args: argparse.Namespace) -> int:
    root: Path = args.root
    hits = 0
    for rec in iter_manifest(root):
        member = rec.get("member", "")
        source = rec.get("source", "")
        if args.member and args.member not in member:
            continue
        if args.source_substr and args.source_substr not in source:
            continue
        print(json.dumps(rec, ensure_ascii=False))
        hits += 1
        if args.limit and hits >= args.limit:
            break
    if hits == 0:
        print("No matches.", file=sys.stderr)
        return 1
    return 0


def cmd_stats(args: argparse.Namespace) -> int:
    """Quick manifest statistics (streams full manifest)."""
    root: Path = args.root
    cats: Counter[str] = Counter()
    shards: Counter[tuple[str, int]] = Counter()
    n = 0
    for rec in iter_manifest(root):
        cats[rec["category"]] += 1
        shards[(rec["category"], rec["shard_id"])] += 1
        n += 1
    print(f"manifest entries: {n:,}")
    for cat, c in cats.most_common():
        print(f"  {cat}: {c:,}")
    if args.verbose:
        for (cat, sid), c in sorted(shards.items()):
            print(f"  {cat} shard {sid:4d}: {c:,}")
    return 0


def build_parser() -> argparse.ArgumentParser:
    parser = argparse.ArgumentParser(
        description=__doc__,
        formatter_class=argparse.RawDescriptionHelpFormatter,
    )
    parser.add_argument(
        "--root",
        type=Path,
        default=DEFAULT_ROOT,
        help=f"Dataset root (default: {DEFAULT_ROOT})",
    )
    sub = parser.add_subparsers(dest="command", required=True)

    p_sum = sub.add_parser("summary", help="Print dataset overview from summary.json")
    p_sum.set_defaults(func=cmd_summary)

    p_ls = sub.add_parser("list-shards", help="List all tar shards")
    p_ls.add_argument("--category", choices=("patent", "product"))
    p_ls.set_defaults(func=cmd_list_shards)

    p_lt = sub.add_parser("list-tar", help="List member names inside one tar")
    p_lt.add_argument("category", choices=("patent", "product"))
    p_lt.add_argument("shard", type=int)
    p_lt.add_argument("--limit", type=int, default=20)
    p_lt.set_defaults(func=cmd_list_tar)

    p_ex = sub.add_parser("extract", help="Extract tar shard(s) to disk")
    p_ex.add_argument("-o", "--output", type=Path, required=True)
    p_ex.add_argument("--category", choices=("patent", "product"))
    p_ex.add_argument("--shard", type=int, help="Single shard id (use with --category)")
    p_ex.add_argument("--flat", action="store_true", help="Do not create category subdirs")
    p_ex.add_argument("--overwrite", action="store_true")
    p_ex.add_argument("--max-files", type=int, help="Stop after N files")
    p_ex.add_argument("--members-file", type=Path, help="Only extract listed basenames")
    p_ex.set_defaults(func=cmd_extract)

    p_sa = sub.add_parser("sample", help="Randomly extract a few images for inspection")
    p_sa.add_argument("-o", "--output", type=Path, required=True)
    p_sa.add_argument("--category", choices=("patent", "product"))
    p_sa.add_argument("--shard", type=int)
    p_sa.add_argument("-n", "--num", type=int, default=8)
    p_sa.add_argument("--seed", type=int, default=42)
    p_sa.add_argument("--pool-size", type=int, default=200_000,
                        help="Max manifest lines to scan before sampling")
    p_sa.add_argument("--flat", action="store_true")
    p_sa.set_defaults(func=cmd_sample)

    p_vf = sub.add_parser("verify", help="Verify tar member counts vs summary.json")
    p_vf.add_argument("--category", choices=("patent", "product"))
    p_vf.add_argument("--shard", type=int)
    p_vf.add_argument("--check-manifest", action="store_true")
    p_vf.set_defaults(func=cmd_verify)

    p_lk = sub.add_parser("lookup", help="Search manifest by member or source substring")
    p_lk.add_argument("--member", help="Substring of tar member name")
    p_lk.add_argument("--source-substr", help="Substring of original source path")
    p_lk.add_argument("--limit", type=int, default=10)
    p_lk.set_defaults(func=cmd_manifest_lookup)

    p_st = sub.add_parser("stats", help="Count manifest entries by category/shard")
    p_st.add_argument("--verbose", action="store_true")
    p_st.set_defaults(func=cmd_stats)

    return parser


def main() -> int:
    parser = build_parser()
    args = parser.parse_args()
    if not args.root.is_dir():
        print(f"Dataset root not found: {args.root}", file=sys.stderr)
        return 1
    return int(args.func(args))


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