CF-GNNExplainer: Counterfactual Explanations for Graph Neural Networks