A local temporal difference code for distributional reinforcement learning