Efficient Deep Imitation and Reinforcement Learning in Multi-agent Enviornments