Fairness API Reference

Fairness Metrics

assert_fairness

High-level function to assert fairness metrics for a model.

from ml_assert.fairness import assert_fairness

# Assert fairness metrics
assert_fairness(
    y_true=y_true,
    y_pred=y_pred,
    sensitive_features=sensitive_features,
    metrics=["demographic_parity", "equal_opportunity"],
    threshold=0.1
)

Parameters

• y_true: True labels
• y_pred: Predicted labels
• sensitive_features: Array of sensitive feature values
• metrics: List of fairness metrics to check
• threshold: Maximum allowed difference between groups (default: 0.1)

Supported Metrics

• demographic_parity: Equal positive prediction rates across groups
• equal_opportunity: Equal true positive rates across groups
• equalized_odds: Equal true positive and false positive rates
• treatment_equality: Equal ratio of false negatives to false positives

Explainability

assert_feature_importance

Assert minimum feature importance scores.

from ml_assert.fairness import assert_feature_importance

# Assert feature importance
assert_feature_importance(
    model=model,
    X=X,
    min_importance=0.1,
    features=["feature1", "feature2"]
)

Parameters

• model: Trained model with feature_importances_ attribute
• X: Feature matrix
• min_importance: Minimum importance score (0.0 to 1.0)
• features: List of features to check (optional)

assert_shap_values

Assert SHAP values for feature importance.

from ml_assert.fairness import assert_shap_values

# Assert SHAP values
assert_shap_values(
    model=model,
    X=X,
    min_importance=0.1,
    features=["feature1", "feature2"]
)

Parameters

• model: Trained model
• X: Feature matrix
• min_importance: Minimum absolute SHAP value
• features: List of features to check (optional)