Computational approach to depression: From the viewpoint of reinforcement learning