On Generating Explanations for Reinforcement Learning Policies: An Empirical Study