Informativeness of Reward Functions in Reinforcement Learning