LANTERN / closedloop /scripts /finalize_benchmark_v1.py
YongshuoLiu's picture
Add LANTERN metadata and closed-loop benchmark
5a5e5cf verified
Raw History Blame Contribute Delete
11.2 kB
#!/usr/bin/env python3
"""Freeze accepted CoDrive routes into a reproducible benchmark catalog."""
from __future__ import annotations
import argparse
import hashlib
import json
import shutil
import statistics
from collections import Counter, defaultdict
from pathlib import Path
PROJECT_ROOT = Path(__file__).resolve().parents[2]
DEFAULT_DATASET_ROOT = PROJECT_ROOT / "closed_dataset"
DEFAULT_RUN_ROOT = DEFAULT_DATASET_ROOT / "acceptance_final_v1_runs"
DEFAULT_OUTPUT = DEFAULT_DATASET_ROOT / "acceptance_final_v1"
def _sha256(path: Path) -> str:
digest = hashlib.sha256()
with path.open("rb") as handle:
for chunk in iter(lambda: handle.read(1024 * 1024), b""):
digest.update(chunk)
return digest.hexdigest()
def _route_path(dataset_root: Path, route_id: str) -> Path:
matches = list((dataset_root / "routes").glob(f"*/{route_id}.xml"))
if len(matches) != 1:
raise RuntimeError(f"expected one XML for {route_id}, found {len(matches)}")
return matches[0]
def _report_path(run_root: Path, route_id: str) -> Path:
matches = list(run_root.glob(f"*/{route_id}/acceptance_report.json"))
if len(matches) != 1:
raise RuntimeError(
f"expected one acceptance report for {route_id}, found {len(matches)}"
)
return matches[0]
def _format_average(values: list[float]) -> str:
return f"{statistics.fmean(values):.2f}" if values else "n/a"
def finalize(dataset_root: Path, run_root: Path, output: Path) -> dict:
manifest_path = dataset_root / "manifest.json"
manifest = json.loads(manifest_path.read_text(encoding="utf-8"))
routes = manifest.get("routes", [])
if len(routes) != 60:
raise RuntimeError(f"expected 60 manifest routes, found {len(routes)}")
output.mkdir(parents=True, exist_ok=True)
reports_dir = output / "reports"
if reports_dir.exists():
shutil.rmtree(reports_dir)
reports_dir.mkdir(parents=True)
scenario_path = dataset_root / "scenarios/closed_coopeval_scenarios.py"
validator_path = dataset_root / "scripts/validate_pdm_acceptance.py"
records = []
errors = []
for route in routes:
route_id = route["route_id"]
route_path = _route_path(dataset_root, route_id)
report_path = _report_path(run_root, route_id)
report = json.loads(report_path.read_text(encoding="utf-8"))
route_hash = _sha256(route_path)
route_errors = []
if not report.get("accepted"):
route_errors.append("acceptance report rejected the route")
if report.get("route_sha256") != route_hash:
route_errors.append("route hash differs from the accepted XML")
if report.get("route_status") not in {"Completed", "Perfect"}:
route_errors.append(f"route status is {report.get('route_status')!r}")
if float(report.get("route_completion", 0.0)) < 99.0:
route_errors.append("route completion is below 99%")
if any(int(value) for value in report.get("collision_counts", {}).values()):
route_errors.append("collision count is nonzero")
if int(report.get("blocked_count", 0)):
route_errors.append("blocked count is nonzero")
if int(report.get("scenario_timeout_count", 0)):
route_errors.append("scenario timeout count is nonzero")
if not report.get("physical_completion_event"):
route_errors.append("physical completion event is missing")
if "hint_on" not in report.get("cooperative_agent_events", []):
route_errors.append("hint_on is missing")
if "hint_off" not in report.get("cooperative_agent_events", []):
route_errors.append("hint_off is missing")
if int(report.get("frame_count", 0)) <= 0:
route_errors.append("no RGB frames were saved")
if int(report.get("video_count", 0)) <= 0:
route_errors.append("no video was saved")
if route_errors:
errors.append({"route_id": route_id, "errors": route_errors})
copied_report = reports_dir / route["family"] / f"{route_id}.json"
copied_report.parent.mkdir(parents=True, exist_ok=True)
shutil.copy2(report_path, copied_report)
records.append(
{
"route_id": route_id,
"family": route["family"],
"town": route["town"],
"source_route_id": route["source_route_id"],
"weather_profile": route["weather_profile"],
"route_length_m": route["route_length_m"],
"route_xml": str(route_path.relative_to(dataset_root)),
"route_sha256": route_hash,
"accepted": not route_errors,
"physical_completion_event": report.get("physical_completion_event"),
"event_release_reason": report.get("event_release_reason"),
"duration_game_s": report.get("duration_game_s"),
"frame_count": report.get("frame_count"),
"route_completion": report.get("route_completion"),
"variation": {
key: route[key]
for key in (
"occluder_model",
"ped_age",
"cut_in_from",
"cut_in_model",
"deterministic_actor",
"obstacle_model",
"car_a_dir",
"ghost_model",
"ghost_speed",
"flow_speed",
)
if key in route
},
"evidence_run_dir": report.get("run_dir"),
}
)
if errors:
raise RuntimeError(json.dumps(errors, indent=2))
family_records: dict[str, list[dict]] = defaultdict(list)
for record in records:
family_records[record["family"]].append(record)
lock = {
"benchmark": "CoDrive",
"version": "1.0",
"route_count": len(records),
"manifest_sha256": _sha256(manifest_path),
"scenario_script_sha256": _sha256(scenario_path),
"acceptance_validator_sha256": _sha256(validator_path),
"source_xml_sha256": manifest["source_xml_sha256"],
"family_counts": dict(Counter(record["family"] for record in records)),
"acceptance": {
"accepted": len(records),
"rejected": 0,
"collision_routes": 0,
"blocked_routes": 0,
"timeout_routes": 0,
},
"routes": records,
}
lock_path = dataset_root / "benchmark_v1.lock.json"
lock_path.write_text(json.dumps(lock, indent=2) + "\n", encoding="utf-8")
lines = [
"# CoDrive Benchmark v1.0 Audit",
"",
"This file is generated by `closed_dataset/scripts/finalize_benchmark_v1.py`.",
"",
"## Release Contract",
"",
f"- Routes: {len(records)} total, 10 per scenario family.",
"- Static validation: passed.",
"- PDM-Lite acceptance: 60/60 passed.",
"- Collisions, blocked routes, and scenario timeouts: 0.",
"- Every run contains hint-on, physical completion, hint-off, RGB frames, and video.",
f"- Scenario implementation SHA256: `{lock['scenario_script_sha256']}`.",
"",
"## Scenario Summary",
"",
"| Scenario family | Routes | Towns | Weather profiles | Mean length (m) | Mean duration (s) |",
"| --- | ---: | ---: | ---: | ---: | ---: |",
]
for family in sorted(family_records):
values = family_records[family]
lines.append(
"| {family} | {routes} | {towns} | {weather} | {length} | {duration} |".format(
family=family.replace("_", " ").title(),
routes=len(values),
towns=len({record["town"] for record in values}),
weather=len({record["weather_profile"] for record in values}),
length=_format_average(
[float(record["route_length_m"]) for record in values]
),
duration=_format_average(
[float(record["duration_game_s"]) for record in values]
),
)
)
lines.extend(
[
"",
"## Reproduction",
"",
"Run the static and PDM-Lite validators before publishing modified routes:",
"",
"```bash",
"python closed_dataset/validate.py --root closed_dataset",
"python closed_dataset/scripts/run_pdm_acceptance_suite.py \\",
" --route-root closed_dataset/routes \\",
" --run-root closed_dataset/acceptance_final_v1_runs \\",
" --gpu 2 --graphics-adapter 0 --port 4897 --tm-port 10897 --rerun",
"python closed_dataset/scripts/finalize_benchmark_v1.py",
"```",
"",
"The exact XML hashes and evidence paths are recorded in",
"`closed_dataset/benchmark_v1.lock.json`.",
"",
]
)
(dataset_root / "BENCHMARK_V1_AUDIT.md").write_text(
"\n".join(lines), encoding="utf-8"
)
catalog_lines = [
"# CoDrive v1.0 Route Catalog",
"",
"| Route | Scenario family | Town | Weather | Length (m) | Source | Variation |",
"| --- | --- | --- | --- | ---: | --- | --- |",
]
for record in sorted(records, key=lambda item: (item["family"], item["route_id"])):
variation = ", ".join(
f"{key}={value}" for key, value in sorted(record["variation"].items())
)
catalog_lines.append(
"| {route} | {family} | {town} | {weather} | {length:.2f} | {source} | {variation} |".format(
route=record["route_id"],
family=record["family"].replace("_", " ").title(),
town=record["town"],
weather=record["weather_profile"].replace("_", " "),
length=float(record["route_length_m"]),
source=record["source_route_id"],
variation=variation or "geometry and actor placement",
)
)
catalog_lines.append("")
(dataset_root / "ROUTE_CATALOG.md").write_text(
"\n".join(catalog_lines), encoding="utf-8"
)
(output / "acceptance_summary.json").write_text(
json.dumps(lock, indent=2) + "\n", encoding="utf-8"
)
return lock
def main() -> int:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--dataset-root", type=Path, default=DEFAULT_DATASET_ROOT)
parser.add_argument("--run-root", type=Path, default=DEFAULT_RUN_ROOT)
parser.add_argument("--output", type=Path, default=DEFAULT_OUTPUT)
args = parser.parse_args()
lock = finalize(
args.dataset_root.resolve(),
args.run_root.resolve(),
args.output.resolve(),
)
print(
f"CoDrive v{lock['version']} frozen: "
f"{lock['acceptance']['accepted']}/{lock['route_count']} accepted"
)
return 0
if __name__ == "__main__":
raise SystemExit(main())