File size: 8,396 Bytes
1137e50 | 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 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 | from __future__ import annotations
import argparse
import json
from pathlib import Path
from typing import Any
PROJECT_ROOT = Path(__file__).resolve().parents[1]
DEFAULT_RESULTS_DIR = PROJECT_ROOT / "eval" / "results"
DEFAULT_OUTPUT_PATH = PROJECT_ROOT / "docs" / "eval_comparison.md"
METRIC_NAMES = [
"json_validity_rate",
"workflow_accuracy",
"status_accuracy",
"required_field_presence_accuracy",
"unsafe_rejection_accuracy",
"false_route_rate",
]
def load_eval_results(results_dir: Path) -> list[dict[str, Any]]:
if not results_dir.exists():
return []
results: list[dict[str, Any]] = []
for path in sorted(results_dir.glob("*.json")):
with path.open("r", encoding="utf-8") as handle:
payload = json.load(handle)
metrics = payload.get("summary_metrics", {})
if not isinstance(metrics, dict):
continue
results.append(
{
"name": _display_name(path, payload),
"path": path,
"metrics": {metric: metrics.get(metric) for metric in METRIC_NAMES},
}
)
return results
def build_markdown_table(results: list[dict[str, Any]]) -> str:
header = "| Model | " + " | ".join(f"`{metric}`" for metric in METRIC_NAMES) + " |"
separator = "| --- | " + " | ".join("---:" for _ in METRIC_NAMES) + " |"
rows = [header, separator]
for result in results:
values = [_format_metric(result["metrics"].get(metric)) for metric in METRIC_NAMES]
rows.append(f"| {result['name']} | " + " | ".join(values) + " |")
return "\n".join(rows)
def build_interpretation(results: list[dict[str, Any]]) -> str:
if not results:
return (
"## Interpretation\n\n"
"No eval result JSON files were found. Run one of the evaluation commands first, "
"then regenerate this comparison report.\n"
)
best_extraction = _best_higher(results, "required_field_presence_accuracy")
safest = _best_safety(results)
false_route = _false_route_summary(results)
next_step = _next_improvement(results)
return (
"## Interpretation\n\n"
f"- Best structured extraction: {best_extraction}.\n"
f"- Safest model: {safest}.\n"
f"- False route rate: {false_route}.\n"
f"- Improve next: {next_step}.\n"
)
def build_report(results: list[dict[str, Any]]) -> str:
table = build_markdown_table(results) if results else "_No eval result files found._"
return (
"# Evaluation Comparison\n\n"
"This report compares RouterCore eval result artifacts from `eval/results/`.\n\n"
"## Metrics\n\n"
f"{table}\n\n"
f"{build_interpretation(results)}"
)
def write_report(
results_dir: Path = DEFAULT_RESULTS_DIR,
output_path: Path = DEFAULT_OUTPUT_PATH,
) -> tuple[Path, list[dict[str, Any]], str]:
results = load_eval_results(results_dir)
report = build_report(results)
output_path.parent.mkdir(parents=True, exist_ok=True)
output_path.write_text(report, encoding="utf-8")
return output_path, results, report
def _display_name(path: Path, payload: dict[str, Any]) -> str:
if "model" in payload:
return str(payload["model"])
if "adapter" in payload:
return f"LoRA: {Path(str(payload['adapter'])).name}"
if path.stem == "fakerouter_eval":
return "FakeRouter"
return path.stem
def _format_metric(value: Any) -> str:
if isinstance(value, (int, float)):
return f"{value:.2%}"
return "n/a"
def _best_higher(results: list[dict[str, Any]], metric: str) -> str:
scored = [
result
for result in results
if isinstance(result["metrics"].get(metric), (int, float))
]
if not scored:
return "not available"
best = max(scored, key=lambda result: result["metrics"][metric])
return f"{best['name']} ({_format_metric(best['metrics'][metric])})"
def _best_lower(results: list[dict[str, Any]], metric: str) -> str:
scored = [
result
for result in results
if isinstance(result["metrics"].get(metric), (int, float))
]
if not scored:
return "not available"
best = min(scored, key=lambda result: result["metrics"][metric])
best_value = best["metrics"][metric]
tied = [result for result in scored if result["metrics"][metric] == best_value]
names = ", ".join(result["name"] for result in tied)
return f"{names} ({_format_metric(best_value)})"
def _best_safety(results: list[dict[str, Any]]) -> str:
scored = [
result
for result in results
if isinstance(result["metrics"].get("unsafe_rejection_accuracy"), (int, float))
and isinstance(result["metrics"].get("false_route_rate"), (int, float))
]
if not scored:
return "not available"
best = max(
scored,
key=lambda result: (
result["metrics"]["unsafe_rejection_accuracy"],
-result["metrics"]["false_route_rate"],
),
)
best_unsafe = best["metrics"]["unsafe_rejection_accuracy"]
best_false_route = best["metrics"]["false_route_rate"]
tied = [
result
for result in scored
if result["metrics"]["unsafe_rejection_accuracy"] == best_unsafe
and result["metrics"]["false_route_rate"] == best_false_route
]
names = ", ".join(result["name"] for result in tied)
label = "models" if len(tied) > 1 else "model"
return (
f"{names} "
f"({label}; unsafe rejection {_format_metric(best_unsafe)}, "
f"false route {_format_metric(best_false_route)})"
)
def _false_route_summary(results: list[dict[str, Any]]) -> str:
best = _best_lower(results, "false_route_rate")
worst_rows = [
result
for result in results
if isinstance(result["metrics"].get("false_route_rate"), (int, float))
and result["metrics"]["false_route_rate"] > 0
]
if not worst_rows:
return f"remained low across available results; best is {best}"
worst = max(worst_rows, key=lambda result: result["metrics"]["false_route_rate"])
return (
f"best is {best}; highest observed is {worst['name']} "
f"({_format_metric(worst['metrics']['false_route_rate'])})"
)
def _next_improvement(results: list[dict[str, Any]]) -> str:
scored = [
result
for result in results
if isinstance(result["metrics"].get("unsafe_rejection_accuracy"), (int, float))
and isinstance(result["metrics"].get("false_route_rate"), (int, float))
and isinstance(result["metrics"].get("required_field_presence_accuracy"), (int, float))
]
if not scored:
return "run at least one evaluation to identify the weakest metric"
safe_candidates = [
result
for result in scored
if result["metrics"]["unsafe_rejection_accuracy"] == 1.0
and result["metrics"]["false_route_rate"] == 0.0
]
candidates = safe_candidates or scored
reference = max(
candidates,
key=lambda result: result["metrics"]["required_field_presence_accuracy"],
)
weaknesses = {
"workflow_accuracy": "workflow classification",
"status_accuracy": "status classification",
"required_field_presence_accuracy": "structured parameter extraction",
"unsafe_rejection_accuracy": "unsafe request rejection",
}
lowest_metric = min(
weaknesses,
key=lambda metric: reference["metrics"].get(metric, 1.0),
)
if reference["metrics"].get("false_route_rate", 0.0) > 0:
return "reduce false routes before optimizing convenience metrics"
return weaknesses[lowest_metric]
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description="Compare RouterCore evaluation result JSON files.")
parser.add_argument("--results-dir", type=Path, default=DEFAULT_RESULTS_DIR)
parser.add_argument("--output", type=Path, default=DEFAULT_OUTPUT_PATH)
return parser.parse_args()
def main() -> None:
args = parse_args()
output_path, results, report = write_report(args.results_dir, args.output)
print(report)
print(f"\nWrote comparison report to {output_path}")
if not results:
print("No result files were found.")
if __name__ == "__main__":
main()
|