Explaining by Imitating: Understanding Decisions by Interpretable Policy Learning