Stable Fitted Reinforcement Learning