A Deep Reinforcement Learning Approach