Implementation of Generative Model.
Note, however, that using grow model in generative network (Generative_v0) still not work as expected. There are still problem with multiple keras.optimizer while calling extend_model(). So this need to manually save and load model when need to upside (e.g. from 16x16 to 32x32). This will be fixed for my next implementation.