Adversial Learning of Prescribed Generative Models