File size: 2,391 Bytes
d6b762b | 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 | """Evaluate a complete set of finite numeric predictions against rational references.
Copyright 2026 Calxo AI. MIT license; see LICENSE.
"""
import argparse
import json
import math
from fractions import Fraction
from pathlib import Path
def load_lines(path):
return [json.loads(line) for line in Path(path).read_text(encoding="utf-8").splitlines() if line.strip()]
def evaluate(reference, predictions):
expected_by_id = {row["case_id"]: row for row in reference}
if len(expected_by_id) != len(reference):
raise ValueError("Reference data contains duplicate IDs.")
actual_by_id = {}
for row in predictions:
case_id, value = row["case_id"], row["prediction"]
if case_id in actual_by_id:
raise ValueError(f"Duplicate prediction: {case_id}")
if isinstance(value, bool) or not isinstance(value, (int, float)) or not math.isfinite(value):
raise ValueError(f"Prediction must be a finite JSON number: {case_id}")
actual_by_id[case_id] = value
missing = set(expected_by_id) - set(actual_by_id)
extra = set(actual_by_id) - set(expected_by_id)
if missing or extra:
raise ValueError(f"Prediction coverage mismatch: {len(missing)} missing, {len(extra)} unknown.")
passed = 0
failures = []
for case_id, row in expected_by_id.items():
expected = float(Fraction(int(row["expected_numerator"]), int(row["expected_denominator"])))
tolerance = max(row["absolute_tolerance"], row["relative_tolerance"] * abs(expected))
if abs(actual_by_id[case_id] - expected) <= tolerance:
passed += 1
else:
failures.append(case_id)
return {"total": len(reference), "predictions": len(predictions), "passed": passed, "failed": len(failures), "accuracy": passed / len(reference) if reference else None, "failed_case_ids": failures}
if __name__ == "__main__":
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("predictions", type=Path)
parser.add_argument("--reference", type=Path, default=Path(__file__).with_name("conversion-reference.jsonl"))
args = parser.parse_args()
try:
report = evaluate(load_lines(args.reference), load_lines(args.predictions))
except (ValueError, KeyError, TypeError, OverflowError) as error:
parser.error(str(error))
print(json.dumps(report, indent=2))
|