"""Unpack PixCell-training-data Parquet shards into the folder layout used by the PixCell training code.

    python to_pixcell_layout.py --hf_dir <snapshot dir> --out <root> [--sources tcga_diagnostic gtex ...]

Writes <root>/patches/<src>/<subtype>/single_1024/<slide_id>/<col>_<row>.jpeg (original JPEG bytes)
and <root>/patches/metadata/patch_names_all.hdf5 with one key per source (e.g. "tcga_diagnostic_1024").
Note: PTRC-HGSOC lives under "others/" (key "others_1024"), matching the original layout.
"""
import argparse
from collections import defaultdict
from pathlib import Path

import h5py
import pyarrow.parquet as pq
from tqdm import tqdm

p = argparse.ArgumentParser()
p.add_argument("--hf_dir", type=Path, required=True)
p.add_argument("--out", type=Path, required=True)
p.add_argument("--sources", nargs="*", default=None)
args = p.parse_args()

patches = args.out / "patches"
names = defaultdict(list)
sources = args.sources or sorted(d.name for d in (args.hf_dir / "data").iterdir())
for source in sources:
    for f in tqdm(sorted((args.hf_dir / "data" / source).glob("*.parquet")), desc=source):
        for batch in pq.ParquetFile(f).iter_batches(columns=["image"], batch_size=256):
            for img in batch.column("image").to_pylist():
                dst = patches / img["path"]
                dst.parent.mkdir(parents=True, exist_ok=True)
                dst.write_bytes(img["bytes"])
                names[img["path"].split("/")[0] + "_1024"].append(img["path"])

(patches / "metadata").mkdir(parents=True, exist_ok=True)
with h5py.File(patches / "metadata/patch_names_all.hdf5", "a") as h5:
    for key, paths in names.items():
        if key in h5:
            del h5[key]
        h5.create_dataset(key, data=paths, dtype=h5py.string_dtype())
print({k: len(v) for k, v in names.items()})