| """ | |
| band_edge_detector/cli.py | |
| ========================= | |
| Command-line interface. | |
| Usage: | |
| band-edge-detector --input scores.csv --output report/ | |
| band-edge-detector --input scores.csv --method pca --n-bins 7 | |
| band-edge-detector --input long_scores.csv --format long | |
| """ | |
| from __future__ import annotations | |
| import argparse | |
| import sys | |
| from pathlib import Path | |
| from .data import BenchmarkMatrix | |
| from .report import analyze | |
| def main(argv=None): | |
| parser = argparse.ArgumentParser( | |
| prog="band-edge-detector", | |
| description="Detect band-edge compression in benchmark scores.", | |
| ) | |
| parser.add_argument("--input", required=True, | |
| help="CSV with models as rows, benchmarks as columns.") | |
| parser.add_argument("--format", choices=["wide", "long"], default="wide", | |
| help="'wide' = models × benchmarks; " | |
| "'long' = (model, benchmark, score).") | |
| parser.add_argument("--output", default="band_edge_report", | |
| help="Directory for figures.") | |
| parser.add_argument("--method", choices=["mean_z", "pca", "irt"], | |
| default="mean_z", | |
| help="Capability proxy method.") | |
| parser.add_argument("--n-bins", type=int, default=5, | |
| help="Quantile bins.") | |
| parser.add_argument("--threshold", type=float, default=0.5, | |
| help="Compression threshold.") | |
| parser.add_argument("--min-coverage", type=float, default=0.5, | |
| help="Drop benchmarks with less coverage than this.") | |
| parser.add_argument("--min-benchmarks", type=int, default=5, | |
| help="Drop models with fewer than this many scores.") | |
| args = parser.parse_args(argv) | |
| if args.format == "wide": | |
| matrix = BenchmarkMatrix.from_csv(args.input) | |
| else: | |
| matrix = BenchmarkMatrix.from_long_csv(args.input) | |
| matrix = matrix.filter_benchmarks(min_coverage=args.min_coverage) | |
| matrix = matrix.filter_models(min_benchmarks=args.min_benchmarks) | |
| print(f"Loaded {matrix.n_models} models × {matrix.n_benchmarks} benchmarks") | |
| report = analyze( | |
| matrix, | |
| method=args.method, | |
| n_bins=args.n_bins, | |
| compression_threshold=args.threshold, | |
| output_dir=args.output, | |
| ) | |
| print(report.summary()) | |
| print(f"\nFigures written to: {args.output}/") | |
| return 0 | |
| if __name__ == "__main__": | |
| sys.exit(main()) |