File size: 3,585 Bytes
c4c86e0 c0b76d1 c4c86e0 c0b76d1 | 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 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 | """Re-run a pull request on MacroSwan/pilot-tasks in simulation and post the score on the PR page.
python scripts/score_pr.py 7 # PR number; needs `pip install -e <cablepilot>[sim,hub]`
python scripts/score_pr.py 7 --seeds 0 1 2 --worlds bedroom playroom --post
The contributor's own result is a claim; this is the check. It downloads the PR's card and tuning (falling back to the
task's merged ones), runs each (world, seed), and compares with the merged best on the same worlds and seeds when the
board has them. Designed to run on the Mac or as a Hugging Face Job (CPU is enough: ~25 min per world and seed).
Nothing from the PR is executed as code: cards and tunings are YAML, validated before the run.
"""
from __future__ import annotations
import argparse
import json
import subprocess
import sys
import tempfile
from pathlib import Path
REPO = "MacroSwan/pilot-tasks"
def main():
ap = argparse.ArgumentParser()
ap.add_argument("pr", type=int)
ap.add_argument("--seeds", type=int, nargs="+", default=[0])
ap.add_argument("--worlds", nargs="+", default=["bedroom"])
ap.add_argument("--post", action="store_true", help="comment the result on the PR (else print it)")
a = ap.parse_args()
from huggingface_hub import HfApi, snapshot_download
api = HfApi()
rev = f"refs/pr/{a.pr}"
details = api.get_discussion_details(REPO, a.pr, repo_type="dataset")
work = Path(tempfile.mkdtemp(prefix=f"pr{a.pr}_"))
snap = Path(snapshot_download(REPO, repo_type="dataset", revision=rev, local_dir=work / "pr"))
tasks = sorted(p.parent.name for p in snap.glob("tasks/*/card.yaml"))
title = details.title
task = next((t for t in tasks if f"[{t}]" in title), None)
if task is None:
sys.exit(f"PR title must start with [<task>]: {title!r}")
tdir = snap / "tasks" / task
card, tuning = tdir / "card.yaml", tdir / "tuning" / "cable.yaml"
lines = [f"**Re-run of PR #{a.pr}** (`{task}`, cable body, simulation)", "",
"| world | seed | done | s / item | furniture ticks | max tension | plan ms |", "|---|---|---|---|---|---|---|"]
rows = []
for w in a.worlds:
for s in a.seeds:
out = work / f"run_{w}_s{s}"
cmd = ["cablepilot", "run", str(card), "--world", w, "--seed", str(s), "--out", str(out), "--by", "pilot-scorer"]
if tuning.exists():
cmd += ["--tuning", str(tuning)]
r = subprocess.run(cmd, capture_output=True, text=True)
if r.returncode != 0 or not (out / "result.json").exists():
lines.append(f"| {w} | {s} | **failed** | | | | |")
print(r.stdout[-2000:], r.stderr[-2000:])
continue
res = json.loads((out / "result.json").read_text())
sc = res["score"]
rows.append(res)
lines.append(f"| {w} | {s} | {sc['done']}/{sc['of']} | {sc['s_per_item']:.0f} | {sc['furniture_contact_ticks']} | "
f"{sc['max_tension_N']:.0f} N | {res['compute']['plan_ms_mean']:.0f} |")
done = sum(r["score"]["done"] for r in rows)
of = sum(r["score"]["of"] for r in rows)
lines += ["", f"Total: **{done}/{of}**. card `{rows[0]['card_sha'] if rows else '?'}`, tuning "
f"`{rows[0]['tuning_sha'] if rows else '?'}`. Same worlds and seeds as the board's best are the fair comparison."]
text = "\n".join(lines)
print(text)
if a.post:
api.comment_discussion(REPO, a.pr, text, repo_type="dataset")
if __name__ == "__main__":
main()
|