When not to use an Adversarial Approach to Generative Modeling