Add standalone conversion script (no lerobot dependency)

convert_to_lerobot_format.py

@@ -0,0 +1,134 @@
+#!/usr/bin/env python
+"""Convert dataset from flat parquet to LeRobot chunked format.
+
+Standalone - no lerobot dependency. Add this to your dataset repo and run:
+
+  python convert_to_lerobot_format.py --data-dir ./data --push-to-hub
+
+Requires: pip install datasets huggingface_hub
+"""
+
+import argparse
+import os
+from pathlib import Path
+
+import numpy as np
+from datasets import Dataset
+from huggingface_hub import delete_file, snapshot_download, upload_folder
+
+CHUNK_FILE_PATTERN = "chunk-{chunk_index:03d}/file-{file_index:03d}"
+DEFAULT_DATA_FILE_SIZE_MB = 100
+DEFAULT_CHUNK_SIZE = 1000
+
+
+def get_dataset_size_mb(ds: Dataset) -> float:
+    return ds.data.nbytes / (1024**2)
+
+
+def update_chunk_file_indices(chunk_idx: int, file_idx: int, chunk_size: int) -> tuple[int, int]:
+    if file_idx == chunk_size - 1:
+        return chunk_idx + 1, 0
+    return chunk_idx, file_idx + 1
+
+
+def convert_flat_to_chunked(data_dir: Path, chunk_size: int = DEFAULT_CHUNK_SIZE) -> bool:
+    flat_files = sorted(data_dir.glob("*.parquet"))
+    chunked_files = list(data_dir.glob("*/*.parquet"))
+
+    if chunked_files:
+        print(f"Already in chunked format ({len(chunked_files)} files).")
+        return False
+
+    if not flat_files:
+        raise FileNotFoundError(f"No parquet files in {data_dir}")
+
+    print(f"Converting {len(flat_files)} flat parquet file(s) to LeRobot chunked format...")
+
+    hf_dataset = Dataset.from_parquet([str(p) for p in flat_files])
+    dataset_size_mb = get_dataset_size_mb(hf_dataset)
+
+    if dataset_size_mb <= DEFAULT_DATA_FILE_SIZE_MB:
+        path = data_dir / f"{CHUNK_FILE_PATTERN.format(chunk_index=0, file_index=0)}.parquet"
+        path.parent.mkdir(parents=True, exist_ok=True)
+        hf_dataset.to_parquet(path)
+        print(f"  Wrote {path}")
+    else:
+        episode_indices = np.array(hf_dataset["episode_index"])
+        episode_boundaries = np.where(np.diff(episode_indices) != 0)[0] + 1
+        episode_starts = np.concatenate(([0], episode_boundaries))
+        episode_ends = np.concatenate((episode_boundaries, [len(hf_dataset)]))
+
+        num_episodes = len(episode_starts)
+        current_episode_idx = 0
+        chunk_idx, file_idx = 0, 0
+
+        while current_episode_idx < num_episodes:
+            shard_start_row = episode_starts[current_episode_idx]
+            shard_end_row = episode_ends[current_episode_idx]
+            next_episode_to_try_idx = current_episode_idx + 1
+
+            while next_episode_to_try_idx < num_episodes:
+                potential_shard_end_row = episode_ends[next_episode_to_try_idx]
+                shard_candidate = hf_dataset.select(
+                    range(shard_start_row, potential_shard_end_row)
+                )
+                if get_dataset_size_mb(shard_candidate) > DEFAULT_DATA_FILE_SIZE_MB:
+                    break
+                shard_end_row = potential_shard_end_row
+                next_episode_to_try_idx += 1
+
+            dataset_shard = hf_dataset.select(range(shard_start_row, shard_end_row))
+            path = data_dir / f"{CHUNK_FILE_PATTERN.format(chunk_index=chunk_idx, file_index=file_idx)}.parquet"
+            path.parent.mkdir(parents=True, exist_ok=True)
+            dataset_shard.to_parquet(path)
+            print(f"  Wrote {path}")
+
+            chunk_idx, file_idx = update_chunk_file_indices(chunk_idx, file_idx, chunk_size)
+            current_episode_idx = next_episode_to_try_idx
+
+    for f in flat_files:
+        f.unlink()
+        print(f"  Removed {f}")
+
+    print("Conversion complete.")
+    return True
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--repo-id", default="ases200q2/libero_object", help="HF dataset repo ID")
+    parser.add_argument("--data-dir", type=Path, required=True, help="Path to data/ directory")
+    parser.add_argument("--push-to-hub", action="store_true")
+    parser.add_argument("--delete-flat-from-hub", action="store_true", help="Remove old flat file from hub")
+    args = parser.parse_args()
+
+    data_dir = Path(args.data_dir)
+    if not data_dir.exists():
+        raise FileNotFoundError(f"Data dir not found: {data_dir}")
+
+    convert_flat_to_chunked(data_dir)
+
+    if args.push_to_hub:
+        root = data_dir.parent
+        print(f"\nPushing to https://huggingface.co/datasets/{args.repo_id} ...")
+        upload_folder(
+            repo_id=args.repo_id,
+            folder_path=str(root),
+            repo_type="dataset",
+            commit_message="Convert data to LeRobot chunked format",
+        )
+        print("Done.")
+
+    if args.delete_flat_from_hub:
+        print(f"\nDeleting flat parquet from {args.repo_id}...")
+        delete_file(
+            path_in_repo="data/train-00000-of-00001.parquet",
+            repo_id=args.repo_id,
+            repo_type="dataset",
+            commit_message="Remove flat parquet",
+        )
+        print("Done.")
+
+
+if __name__ == "__main__":
+    main()