Reinforcement Learning with Multiple Experts: A Bayesian Model Combination Approach