Learning Sparse Representations in Reinforcement Learning with Sparse Coding