Files
yao-meta-skill/scripts/render_eval_dashboard.py

127 lines
4.5 KiB
Python

#!/usr/bin/env python3
import argparse
import json
import subprocess
import sys
from pathlib import Path
BEGIN_MARKER = "<!-- BEGIN:EVAL_RESULTS -->"
END_MARKER = "<!-- END:EVAL_RESULTS -->"
def run_eval_suite(args: argparse.Namespace) -> dict:
proc = subprocess.run(
[
sys.executable,
"scripts/run_eval_suite.py",
"--eval-dir",
args.eval_dir,
"--description-file",
args.description_file,
"--baseline-description-file",
args.baseline_description_file,
],
capture_output=True,
text=True,
check=True,
)
return json.loads(proc.stdout)
def render_readme_panel(report: dict) -> str:
summary = report["summary"]
suites = report["suites"]
families = report["family_summary"]
lines = [
BEGIN_MARKER,
f"- regression corpus: `{summary['total_cases']}` prompts across `{summary['family_count']}` families",
f"- aggregate result: `{summary['false_positives']}` false positives, `{summary['false_negatives']}` false negatives, average precision `{summary['average_precision']}`, average recall `{summary['average_recall']}`",
"- suite status:",
"",
"| Suite | Cases | FP | FN | Precision | Recall |",
"| --- | ---: | ---: | ---: | ---: | ---: |",
]
for suite_name in ("train", "dev", "holdout"):
suite = suites[suite_name]
cases = sum(bucket["total"] for bucket in suite["bucket_stats"].values())
lines.append(
f"| {suite_name} | {cases} | {suite['false_positives']} | {suite['false_negatives']} | {suite['precision']} | {suite['recall']} |"
)
lines.extend(
[
"",
"| Family | Cases | Pass Rate |",
"| --- | ---: | ---: |",
]
)
for family, stats in sorted(families.items()):
lines.append(f"| `{family}` | {stats['total']} | {stats['pass_rate']} |")
lines.extend(
[
"",
"Full reports: [reports/eval_suite.json](reports/eval_suite.json) and [reports/family_summary.md](reports/family_summary.md)",
END_MARKER,
]
)
return "\n".join(lines)
def render_family_summary(report: dict) -> str:
summary = report["summary"]
lines = [
"# Eval Family Summary",
"",
f"- total cases: `{summary['total_cases']}`",
f"- families: `{summary['family_count']}`",
f"- false positives: `{summary['false_positives']}`",
f"- false negatives: `{summary['false_negatives']}`",
f"- average precision: `{summary['average_precision']}`",
f"- average recall: `{summary['average_recall']}`",
"",
"## Family Results",
"",
"| Family | Cases | Passed | FP | FN | Pass Rate |",
"| --- | ---: | ---: | ---: | ---: | ---: |",
]
for family, stats in sorted(report["family_summary"].items()):
lines.append(
f"| `{family}` | {stats['total']} | {stats['passed']} | {stats['false_positives']} | {stats['false_negatives']} | {stats['pass_rate']} |"
)
return "\n".join(lines) + "\n"
def update_readme(readme_path: Path, panel: str) -> None:
text = readme_path.read_text(encoding="utf-8")
if BEGIN_MARKER not in text or END_MARKER not in text:
raise SystemExit(f"README markers not found in {readme_path}")
start = text.index(BEGIN_MARKER)
end = text.index(END_MARKER) + len(END_MARKER)
updated = text[:start] + panel + text[end:]
readme_path.write_text(updated, encoding="utf-8")
def main() -> None:
parser = argparse.ArgumentParser(description="Render README-visible eval results and full family report.")
parser.add_argument("--eval-dir", default="evals")
parser.add_argument("--description-file", default="evals/improved_description.txt")
parser.add_argument("--baseline-description-file", default="evals/baseline_description.txt")
parser.add_argument("--output-dir", default="reports")
parser.add_argument("--readme", default="README.md")
args = parser.parse_args()
report = run_eval_suite(args)
output_dir = Path(args.output_dir)
output_dir.mkdir(parents=True, exist_ok=True)
(output_dir / "eval_suite.json").write_text(json.dumps(report, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
(output_dir / "family_summary.md").write_text(render_family_summary(report), encoding="utf-8")
update_readme(Path(args.readme), render_readme_panel(report))
if __name__ == "__main__":
main()