peal.explainers

Explainers: methods that explain a predictor’s decisions.

The counterfactual explainers in counterfactual_explainer (SCE, ACE, TIME, DAE-distill and the perfect-false-counterfactual oracle) pair a predictor with a generator from peal.generators and search for minimal edits that flip the prediction; no_generator_counterfactual_explainers (DiCE) works on tabular data without a generator and lrp_explainer yields attribution heatmaps. explainer_factory.get_explainer instantiates one from an ExplainerConfig; adaptors such as CFKD call explain_batch to collect the explanations a teacher then labels.

Submodules

counterfactual_explainer	The counterfactual explainer of PEAL and the configs of its baselines.
explainer_factory	Factory that resolves an explainer specification into an ExplainerInterface.
interfaces	Base config and interface shared by all PEAL explainers.
lrp_explainer	Attribution (heatmap) explainer built on zennit.
no_generator_counterfactual_explainers	Counterfactual explainers that need no PEAL generator.