Reinforcement Learning in the Wild with Maximum Likelihood-based Model Transfer