peal.explainers.lrp_explainer¶

Attribution (heatmap) explainer built on zennit.

Besides counterfactuals PEAL can explain a classifier with pixel-wise attribution maps. LRPExplainer wraps zennit’s IntegratedGradients attributor with an EpsilonGammaBox LRP composite and optional model canonizers, and exposes the same explain_batch/run interface the counterfactual explainers use, so it can be plugged into the visualization tools (peal.visualization.model_comparison).

Classes