Post hoc explanations approximate the behavior of a black-box by extracting relationships between feature values and the predictions. Several local explanation methods are model-agnostic, meaning they do not have access to the internal structure of the model.
