Generative causal explanations of black-box classifiers