"""Build Dataset Viewer parquet shards: 1024px JPEG previews + per-image counts."""
import io
from pathlib import Path

import pandas as pd
from datasets import Dataset, Features, Image as HFImage, Value
from PIL import Image, ImageFile
from tqdm import tqdm

# 3 source images are truncated (1597915859896.jpg, 1599030751681.jpg,
# 1599042714933.jpg); previews render the decodable portion instead of crashing.
ImageFile.LOAD_TRUNCATED_IMAGES = True

ROOT = Path(__file__).resolve().parents[1]
DATA = ROOT / "data"
OUT = ROOT / "parquet"
SHARDS = {"train": 4, "val": 1, "test": 1}
FEATURES = Features(
    {"image": HFImage(), "berry_count": Value("int32"), "cluster_count": Value("int32")}
)


def preview_bytes(path):
    img = Image.open(path).convert("RGB")
    img.thumbnail((1024, 1024))
    buf = io.BytesIO()
    img.save(buf, format="JPEG", quality=85)
    return buf.getvalue()


def main():
    OUT.mkdir(exist_ok=True)
    meta = pd.read_csv(DATA / "metadata.csv")
    for split, n_shards in SHARDS.items():
        rows = meta[meta.split == split].reset_index(drop=True)

        def gen(rows=rows):
            for r in tqdm(rows.itertuples(), total=len(rows), desc=split):
                yield {
                    "image": {"bytes": preview_bytes(DATA / "imgs" / r.file_name),
                              "path": r.file_name},
                    "berry_count": int(r.berry_count),
                    "cluster_count": int(r.cluster_count),
                }

        ds = Dataset.from_generator(gen, features=FEATURES)
        for i in range(n_shards):
            shard_path = OUT / f"{split}-{i:05d}-of-{n_shards:05d}.parquet"
            ds.shard(num_shards=n_shards, index=i).to_parquet(shard_path)
            print("wrote", shard_path, f"{shard_path.stat().st_size/1e6:.0f} MB")


if __name__ == "__main__":
    main()