Batch Reinforcement Learning Theoretical Comparison of Q Approximation Schemes - arxiv.org