Evaluating the Performance of Reinforcement Learning Algorithms