A Bayesian Approach to Generative Adversarial Imitation Learning