File size: 2,064 Bytes
9ffd3e8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 | """Merge phase 2 datasets generated separately (different seeds) into one.
Uses LeRobot's aggregate_datasets, then joins the per-episode notes
(phase2_episodes.jsonl, renumbered in merged order) and the generation summaries.
Run: .venv/bin/python sim/merge_datasets.py data/so101_chess_sim_all data/so101_chess_sim data/so101_chess_sim_b
"""
from __future__ import annotations
import argparse
import json
from pathlib import Path
def main():
from lerobot.datasets.aggregate import aggregate_datasets
from lerobot.datasets.lerobot_dataset import LeRobotDataset
ap = argparse.ArgumentParser(description=__doc__)
ap.add_argument("out")
ap.add_argument("parts", nargs="+")
args = ap.parse_args()
parts = [Path(p) for p in args.parts]
out = Path(args.out)
aggregate_datasets(repo_ids=[f"local/part{i}" for i in range(len(parts))], aggr_repo_id="local/so101_chess_sim",
roots=parts, aggr_root=out)
offset, lines, summaries = 0, [], []
for p in parts:
recs = [json.loads(line) for line in (p / "phase2_episodes.jsonl").read_text().splitlines()]
for r in recs:
r["part"] = p.name
r["episode_index"] = offset + r["episode_index"]
lines.append(json.dumps(r))
offset += len(recs)
summaries.append(json.loads((p / "phase2_summary.json").read_text()))
(out / "phase2_episodes.jsonl").write_text("\n".join(lines) + "\n")
total = dict(episodes=sum(s["episodes"] for s in summaries), attempts=sum(s["attempts"] for s in summaries),
minutes=round(sum(s["minutes"] for s in summaries), 1), parts=[p.name for p in parts],
config=summaries[-1].get("config"))
total["success_rate"] = round(total["episodes"] / total["attempts"], 4)
(out / "phase2_summary.json").write_text(json.dumps(total, indent=2))
ds = LeRobotDataset("local/so101_chess_sim", root=out)
print(f"{out}: {ds.num_episodes} episodes, {ds.num_frames} frames; notes {len(lines)}")
if __name__ == "__main__":
main()
|