Fairlearn
Fairness metrics + mitigation. Demographic parity, equalized odds. Mitigation via reweighting or threshold optimization.
Advertisement
InterpretML
SHAP, LIME, glass-box models. Explanations for individual predictions + global model behavior.
Advertisement
Error Analysis
Cohort-level error breakdown. Find where model fails. Compare across groups.
Counterfactual analysis
DiCE: generate counterfactuals ('what would change prediction?'). Fairness + interpretability.