interpretml/interpret

Combines glassbox models (EBM, decision trees, linear models) with post-hoc explainers (SHAP, LIME, partial dependence) in a unified API. Features Explainable Boosting Machines that match state-of-the-art blackbox performance while remaining fully interpretable with automatic interaction detection and differential privacy support. Integrates with scikit-learn ecosystems and provides Plotly/Dash-based dashboards for both global and local explanations across multiple models.

6,813 stars. Actively maintained with 74 commits in the last 30 days.