from __future__ import annotations import argparse import logging from pathlib import Path from huggingface_hub import HfFileSystem, hf_hub_download DATASET_FILES: dict[str, tuple[str, ...]] = { "talkpl-ai/TalkPlayData-Challenge-User-Metadata": ( "data/all_users-00000-of-00001.parquet", ), "talkpl-ai/TalkPlayData-Challenge-Track-Metadata": ( "data/all_tracks-00000-of-00001.parquet", "data/test_tracks-00000-of-00001.parquet", ), "talkpl-ai/TalkPlayData-Challenge-User-Embeddings": ( "data/train-00000-of-00001.parquet", "data/test_warm-00000-of-00001.parquet", "data/test_cold-00000-of-00001.parquet", ), "talkpl-ai/TalkPlayData-Challenge-Track-Embeddings": ( "data/all_tracks-*.parquet", "data/test_tracks-00000-of-00001.parquet", ), "talkpl-ai/TalkPlayData-Challenge-Dataset": ( "data/train-00000-of-00001.parquet", "data/test-00000-of-00001.parquet", ), "talkpl-ai/TalkPlayData-Challenge-Blind-A": ( "data/test-00000-of-00001.parquet", ), } def parse_args() -> argparse.Namespace: parser = argparse.ArgumentParser(description="Download TalkPlay datasets from Hugging Face") parser.add_argument( "--output-dir", type=Path, default=Path(__file__).resolve().parents[3] / "data", help="Local output directory (default: project_root/data)", ) parser.add_argument( "--overwrite", action="store_true", help="Re-download files even if they already exist locally", ) parser.add_argument( "--dry-run", action="store_true", help="Show which files would be downloaded without downloading", ) return parser.parse_args() def expand_patterns(repo_id: str, patterns: tuple[str, ...], fs: HfFileSystem) -> list[str]: files: list[str] = [] for pattern in patterns: if "*" in pattern or "?" in pattern or "[" in pattern: remote_pattern = f"datasets/{repo_id}/{pattern}" matches = fs.glob(remote_pattern) prefix = f"datasets/{repo_id}/" files.extend(path.removeprefix(prefix) for path in sorted(matches)) else: files.append(pattern) return files def download_repo_files( repo_id: str, files: list[str], output_root: Path, overwrite: bool, dry_run: bool, ) -> tuple[int, int]: downloaded = 0 skipped = 0 repo_output_dir = output_root / repo_id for relative_file in files: destination = repo_output_dir / relative_file if destination.exists() and destination.stat().st_size > 0 and not overwrite: logging.info("SKIP %s", destination) skipped += 1 continue destination.parent.mkdir(parents=True, exist_ok=True) if dry_run: logging.info("DRY-RUN download %s -> %s", f"{repo_id}/{relative_file}", destination) continue logging.info("DOWNLOAD %s", f"{repo_id}/{relative_file}") hf_hub_download( repo_id=repo_id, repo_type="dataset", filename=relative_file, local_dir=repo_output_dir, force_download=overwrite, ) downloaded += 1 return downloaded, skipped def main() -> None: args = parse_args() logging.basicConfig(level=logging.INFO, format="%(levelname)s - %(message)s") logging.getLogger("httpx").setLevel(logging.WARNING) fs = HfFileSystem() total_downloaded = 0 total_skipped = 0 for repo_id, patterns in DATASET_FILES.items(): files = expand_patterns(repo_id, patterns, fs) if not files: logging.warning("No files matched for repository %s", repo_id) continue downloaded, skipped = download_repo_files( repo_id=repo_id, files=files, output_root=args.output_dir, overwrite=args.overwrite, dry_run=args.dry_run, ) total_downloaded += downloaded total_skipped += skipped logging.info( "Completed. Downloaded=%s, Skipped=%s, Output=%s", total_downloaded, total_skipped, args.output_dir, ) if __name__ == "__main__": main()