test-int/compare_search_benchmarks.py
Rama main · réplica de lectura del repositorio original.
#!/usr/bin/env python3
"""Compare two search benchmark JSONL files and report metric deltas."""
from __future__ import annotations
import argparse
import json
from dataclasses import dataclass
from pathlib import Path
from typing import Iterable
LOWER_IS_BETTER_SUFFIXES = ("_ms", "_seconds", "_size_mb", "_size_bytes")
HIGHER_IS_BETTER_SUFFIXES = ("_per_sec",)
HIGHER_IS_BETTER_PREFIXES = ("hit_rate_", "recall_", "mrr_")
EQUAL_IS_BETTER_KEYS = {"notes_indexed", "queries_executed"}
LOWER_IS_BETTER_KEYS = {
"accepted_empty_rate",
"negative_false_positive_rate",
"wrong_top_rate",
"model_cache_bytes",
"peak_rss_bytes",
"rss_after_index_bytes",
"rss_after_load_bytes",
"rss_model_delta_bytes",
"vector_storage_bytes",
}
@dataclass(frozen=True)
class BenchmarkRecord:
benchmark: str
metrics: dict[str, float]
timestamp_utc: str | None
def _preference_for_metric(metric_name: str) -> str:
"""Return optimization preference for a metric."""
if metric_name in EQUAL_IS_BETTER_KEYS:
return "equal"
if metric_name in LOWER_IS_BETTER_KEYS:
return "lower"
if metric_name.startswith(HIGHER_IS_BETTER_PREFIXES):
return "higher"
if metric_name.endswith(HIGHER_IS_BETTER_SUFFIXES):
return "higher"
if metric_name.endswith(LOWER_IS_BETTER_SUFFIXES):
return "lower"
return "none"
def _classify_delta(metric_name: str, baseline: float, candidate: float) -> str:
"""Classify candidate metric movement relative to baseline."""
if candidate == baseline:
return "same"
preference = _preference_for_metric(metric_name)
if preference == "higher":
return "better" if candidate > baseline else "worse"
if preference == "lower":
return "better" if candidate < baseline else "worse"
if preference == "equal":
return "better" if candidate == baseline else "worse"
return "n/a"
def _format_delta_percent(baseline: float, delta: float) -> str:
if baseline == 0:
return "n/a"
return f"{(delta / baseline) * 100:+.2f}%"
def _read_latest_records(path: Path) -> dict[str, BenchmarkRecord]:
records: dict[str, BenchmarkRecord] = {}
with path.open("r", encoding="utf-8") as file:
for line_number, line in enumerate(file, start=1):
stripped = line.strip()
if not stripped:
continue
try:
payload = json.loads(stripped)
except json.JSONDecodeError as exc: # pragma: no cover - invalid file input path
raise ValueError(f"{path}:{line_number}: invalid JSON ({exc})") from exc
benchmark = payload.get("benchmark")
metrics = payload.get("metrics")
timestamp_utc = payload.get("timestamp_utc")
if not isinstance(benchmark, str):
raise ValueError(f"{path}:{line_number}: missing or invalid 'benchmark'")
if not isinstance(metrics, dict):
raise ValueError(f"{path}:{line_number}: missing or invalid 'metrics'")
if timestamp_utc is not None and not isinstance(timestamp_utc, str):
raise ValueError(f"{path}:{line_number}: invalid 'timestamp_utc'")
numeric_metrics: dict[str, float] = {}
for metric_name, metric_value in metrics.items():
if isinstance(metric_value, bool):
continue
if isinstance(metric_value, (int, float)):
numeric_metrics[str(metric_name)] = float(metric_value)
records[benchmark] = BenchmarkRecord(
benchmark=benchmark,
metrics=numeric_metrics,
timestamp_utc=timestamp_utc,
)
return records
def _iter_rows(
baseline_records: dict[str, BenchmarkRecord],
candidate_records: dict[str, BenchmarkRecord],
include_benchmarks: set[str] | None = None,
) -> Iterable[list[str]]:
common_benchmarks = sorted(set(baseline_records).intersection(candidate_records))
if include_benchmarks:
common_benchmarks = [name for name in common_benchmarks if name in include_benchmarks]
for benchmark in common_benchmarks:
baseline = baseline_records[benchmark]
candidate = candidate_records[benchmark]
common_metrics = sorted(set(baseline.metrics).intersection(candidate.metrics))
for metric in common_metrics:
baseline_value = baseline.metrics[metric]
candidate_value = candidate.metrics[metric]
delta = candidate_value - baseline_value
yield [
benchmark,
metric,
f"{baseline_value:.6f}",
f"{candidate_value:.6f}",
f"{delta:+.6f}",
_format_delta_percent(baseline_value, delta),
_classify_delta(metric, baseline_value, candidate_value),
]
def _print_table(headers: list[str], rows: list[list[str]]) -> None:
all_rows = [headers, *rows]
widths = [max(len(row[index]) for row in all_rows) for index in range(len(headers))]
def format_row(row: list[str]) -> str:
return " | ".join(value.ljust(widths[index]) for index, value in enumerate(row))
print(format_row(headers))
print("-+-".join("-" * width for width in widths))
for row in rows:
print(format_row(row))
def _print_markdown_table(headers: list[str], rows: list[list[str]]) -> None:
print("| " + " | ".join(headers) + " |")
print("| " + " | ".join(["---"] * len(headers)) + " |")
for row in rows:
print("| " + " | ".join(row) + " |")
def _print_missing(
baseline_records: dict[str, BenchmarkRecord],
candidate_records: dict[str, BenchmarkRecord],
) -> None:
baseline_only = sorted(set(baseline_records) - set(candidate_records))
candidate_only = sorted(set(candidate_records) - set(baseline_records))
if baseline_only:
print("\nBenchmarks only in baseline:")
for benchmark in baseline_only:
print(f"- {benchmark}")
if candidate_only:
print("\nBenchmarks only in candidate:")
for benchmark in candidate_only:
print(f"- {benchmark}")
def main() -> int:
parser = argparse.ArgumentParser(
description="Compare two search benchmark JSONL outputs and print metric deltas."
)
parser.add_argument("baseline", type=Path, help="Path to baseline benchmark JSONL file")
parser.add_argument("candidate", type=Path, help="Path to candidate benchmark JSONL file")
parser.add_argument(
"--benchmarks",
type=str,
default="",
help="Comma-separated benchmark names to include (default: all common benchmarks)",
)
parser.add_argument(
"--show-missing",
action="store_true",
help="Print benchmark names present in only one file",
)
parser.add_argument(
"--format",
choices=("table", "markdown"),
default="table",
help="Output format for comparison rows",
)
args = parser.parse_args()
if not args.baseline.exists():
raise SystemExit(f"Baseline file not found: {args.baseline}")
if not args.candidate.exists():
raise SystemExit(f"Candidate file not found: {args.candidate}")
baseline_records = _read_latest_records(args.baseline)
candidate_records = _read_latest_records(args.candidate)
include_benchmarks = {
benchmark.strip()
for benchmark in args.benchmarks.split(",")
if benchmark and benchmark.strip()
}
if not include_benchmarks:
include_benchmarks = None
rows = list(
_iter_rows(
baseline_records=baseline_records,
candidate_records=candidate_records,
include_benchmarks=include_benchmarks,
)
)
if not rows:
print("No comparable benchmark metrics found.")
else:
headers = ["benchmark", "metric", "baseline", "candidate", "delta", "delta_pct", "status"]
if args.format == "markdown":
_print_markdown_table(headers=headers, rows=rows)
else:
_print_table(headers=headers, rows=rows)
if args.show_missing:
_print_missing(baseline_records, candidate_records)
return 0
if __name__ == "__main__":
raise SystemExit(main())