| """Verify uploaded fold weights against one stored OOF prediction per fold.""" |
|
|
| from __future__ import annotations |
|
|
| import csv |
| import json |
| from pathlib import Path |
|
|
| import numpy as np |
|
|
| from inference import PseudoSnpSVD64 |
|
|
|
|
| ROOT = Path(__file__).resolve().parent |
| OOF = ROOT / "results" / "pseudosnp" / "oof_predictions_pip_ge_0.30.tsv" |
|
|
|
|
| def main() -> None: |
| model = PseudoSnpSVD64(ROOT) |
| examples: dict[int, dict[str, str]] = {} |
| with OOF.open("r", encoding="utf-8", newline="") as handle: |
| for row in csv.DictReader(handle, delimiter="\t"): |
| fold = int(row["outer_fold"]) |
| examples.setdefault(fold, row) |
| if len(examples) == 5: |
| break |
|
|
| if set(examples) != set(range(5)): |
| raise RuntimeError("stored OOF evidence does not cover all five folds") |
|
|
| checks = [] |
| for fold in range(5): |
| row = examples[fold] |
| observed = float( |
| model.score_pair_by_fold(row["sequence_ref"], row["sequence_alt"])[fold] |
| ) |
| expected = float(row["transfer_fixed_raw_score"]) |
| close = bool(np.isclose(observed, expected, rtol=0.0, atol=5e-6)) |
| checks.append( |
| { |
| "fold": fold, |
| "example_id": row["example_id"], |
| "expected": expected, |
| "observed": observed, |
| "absolute_error": abs(observed - expected), |
| "passed": close, |
| } |
| ) |
| if not all(item["passed"] for item in checks): |
| raise AssertionError(json.dumps(checks, indent=2)) |
| print(json.dumps({"status": "PASS", "checks": checks}, indent=2)) |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
