Interpretable Counterfactual Explanations Guided by Prototypes

We propose a fast, model agnostic method for finding interpretable counterfactual explanations of classifier predictions by using class prototypes. show that prototypes, obtained either an encoder or through specific k-d trees, significantly speed up the search instances and result in more explanations. quantitatively evaluate interpretability generated counterfactuals to illustrate effectiveness our on image tabular dataset, respectively MNIST Breast Cancer Wisconsin (Diagnostic). Additionally, we principled approach handle categorical variables Adult (Census) dataset. Our also eliminates computational bottleneck arises because numerical gradient evaluation black box models.