Emulating Doctors ' Reasoning with Deep Reinforcement Learning