Towards The Understanding Of Sample Efficient Reinforcement Learning Algorithms