Testing Python (ML) active any

Feature extraction failures fail fast instead of degrading silently

test-python-ml-004

Intent

Feature extractors should fail loudly when required inputs or dependencies are broken, rather than silently returning NaNs, nulls, empty values, or partial outputs that corrupt model behavior.

Applicability

Applies when the diff changes feature extraction, preprocessing validation, or test or gate logic around extracted features. Return unknown when the code does not affect feature extraction behavior.

What to inspect

Check extractor outputs, validation assertions, NaN or null handling, dependency failures, and whether tests or gates treat broken extraction as a hard failure.

Pass criteria

Broken extraction paths are detected and turned into explicit failures through assertions, validators, or tests, rather than silently flowing through as ordinary feature values.

Fail criteria

The changed extractor swallows dependency failures, returns silent NaNs or nulls for required features, or lets partial feature sets proceed with no hard failure.

Do not flag

Optional best-effort enrichments that are clearly marked optional, or legitimate zero-valued features that are not acting as hidden error sentinels.

Confidence guidance

HIGH when silent invalid feature outputs are directly visible. MEDIUM when helpers hide part of the extraction path. LOW when optionality is unclear.

Remediation

Add explicit validation and tests that fail when required extracted features are missing, invalid, or partial.

Pass example

features = extract_features(record)
if np.isnan(features['age_bucket_score']):
    raise ValueError('feature extraction failed')

Fail example

features = extract_features(record)
features['age_bucket_score'] = np.nan_to_num(features['age_bucket_score'])
model.predict([features])