Inverse Reinforcement Learning Meets Power Allocation in Multi-user Cellular Networks