Explainable Reinforcement Learning via Model Transforms