Learning Representations In Reinforcement Learning