Task-Guided Inverse Reinforcement Learning Under Partial Information