Counterfactual Explanations Of Learned Reward Functions