Characterizing reinforcement learning methods through parameterized learning problems