Regularized Softmax Deep Multi-Agent Q-Learning