Imitating Opponent to Win: Adversarial Policy Imitation Learning in Two-player Competitive Games