Coordinating Multi-Agent Deep Reinforcement Learning in Wargame