Local Model-Agnostic Methods
Techniques for explaining individual predictions across any model.
Chapter 7: Ceteris Paribus and ICE Plots Ceteris Paribus profiles, Individual Conditional Expectation (ICE) [Feature sensitivity, conditional analysis, visualization] Chapter 8: LIME and Anchors Local Interpretable Model-agnostic Explanations (LIME), Scoped Rules (Anchors) [Local surrogates, rule-based explanations, stability] Chapter 9: Counterfactual Explanations Generating actionable what-if scenarios [Nearest counterfactuals, plausibility constraints, optimization] Chapter 10: SHAP and Shapley Values Shapley Additive Explanations (SHAP), underlying Shapley value theory [Additive feature attribution, TreeSHAP, KernelSHAP] Chapter 11: Scalable Local Explanations Efficient methods for large-scale and trillion-parameter models [FastSHAP, approximate counterfactuals, sampling-based SHAP, gradient checkpointing]