Inverse Reinforcement Learning from Non-Stationary Learning Agents