DriveSTA / scripts /read_trajectories.py
Rtwotwo's picture
Release trajectory shards, raw checkpoints, scene inventories and recovery tools
0da2563
Raw History Blame Contribute Delete
1.61 kB
"""Read a complete scene bank with NumPy mmap and optional high-score targets."""
import argparse
import json
from pathlib import Path
import numpy as np
def read_scene(directory):
"""Return an immutable mmap; COMPLETE may describe only the high-score bank."""
directory = Path(directory)
completion = json.loads((directory / "ALL_SCORES.json").read_text())
if not completion.get("passed"):
raise ValueError("The full candidate bank has not passed its source checks")
rows = np.load(directory / "candidates.npy", mmap_mode="r", allow_pickle=False)
count = completion["candidates"]
if count > len(rows) or not {"trajectories", "labels"}.issubset(rows.dtype.names or ()):
raise ValueError("Invalid candidate bank schema")
return rows[:count]
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("scene", type=Path, help="Extracted scenes/train/<token> directory")
args = parser.parse_args()
rows = read_scene(args.scene)
report = {
"scene": str(args.scene),
"candidates": len(rows),
"trajectory_shape": list(rows["trajectories"].shape),
"label_shape": list(rows["labels"].shape),
"dtype": str(rows.dtype),
"mean_pdms": float(rows["labels"][:, 6].mean()),
}
high = args.scene / "high_score.npz"
if high.exists():
with np.load(high, allow_pickle=False) as bank:
report["high_score_fields"] = {key: list(bank[key].shape) for key in bank.files}
print(json.dumps(report, indent=2))
if __name__ == "__main__":
main()