Review for NeurIPS paper: Generative causal explanations of black-box classifiers