Reinforcement Learning with Non-Cumulative Objective