Datasets:
Download scripts/tune_trackD.py from Nabidnur/RoboRACE-Bench: direct link, hf CLI and curl.
- Browser
- Download file 2.26 kB
-
https://huggingface.co/datasets/Nabidnur/RoboRACE-Bench/resolve/main/scripts/tune_trackD.py
- Command line
-
hf download hf://datasets/Nabidnur/RoboRACE-Bench/scripts/tune_trackD.py
-
curl -L -o tune_trackD.py https://huggingface.co/datasets/Nabidnur/RoboRACE-Bench/resolve/main/scripts/tune_trackD.py
2.26 kB
| #!/usr/bin/env python3 | |
| """Development-set tuning for Track D (mazes with seeds 0-39; the benchmark uses 1000-1019). | |
| python scripts/tune_trackD.py # compares fusion rules and risk-planner settings on dev mazes | |
| Writes results/trackD/dev_tuning.json. Nothing here touches the benchmark seeds. | |
| """ | |
| import itertools | |
| import json | |
| import sys | |
| from multiprocessing import Pool | |
| from pathlib import Path | |
| import numpy as np | |
| sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "src")) | |
| from robotics.bench import trackD # noqa: E402 | |
| def _job(args): | |
| seed, attack, F, method, tau = args | |
| if tau is not None: | |
| trackD.NET_TAU = tau | |
| return trackD.episode(seed, attack, F, method)["spl"] | |
| def score(method, seeds, tau=None, pool=None): | |
| jobs = [(s, a, F, method, tau) for a in trackD.ATTACKS_D for F in ((0,) if a == "none" else (1, 2, 3, 4)) for s in seeds] | |
| spl = pool.map(_job, jobs, chunksize=8) | |
| by = {} | |
| for (s, a, F, *_), v in zip(jobs, spl): | |
| by.setdefault(a, []).append(v) | |
| return float(np.mean(spl)), {a: round(100 * float(np.mean(v)), 1) for a, v in by.items()} | |
| def main(): | |
| n = int(sys.argv[1]) if len(sys.argv) > 1 else 12 | |
| only = sys.argv[2].split(",") if len(sys.argv) > 2 else None | |
| seeds = list(range(n)) | |
| out = {"dev_seeds": [0, n - 1], "methods": {}, "grid": {}} | |
| with Pool(4) as pool: | |
| for m in trackD.available_methods(): | |
| if only is None or "methods" in only: | |
| mean, by = score(m, seeds, pool=pool) | |
| out["methods"][m] = {"mean": mean, **by} | |
| print(f"{m:30s} {100 * mean:5.1f}", by, flush=True) | |
| if only is None or "grid" in only: | |
| for m, tau in itertools.product(("race_net_m", "race_net_m" + trackD.MV), (1.0, 2.0, 3.0)): | |
| mean, by = score(m, seeds, tau, pool) | |
| out["grid"][f"{m}|tau={tau}"] = {"mean": mean, **by} | |
| print(f"{m} tau {tau}: {100 * mean:5.1f}", by, flush=True) | |
| f = Path("results/trackD/dev_tuning.json") | |
| prev = json.loads(f.read_text()) if f.exists() else {} | |
| for k in ("methods", "grid"): | |
| out[k] = {**prev.get(k, {}), **out[k]} | |
| f.write_text(json.dumps(out, indent=1)) | |
| if __name__ == "__main__": | |
| main() | |