Deep reinforcement learning to multi-agent deep reinforcement learning