Learning Strategy Representation for Imitation Learning in Multi-Agent Games