C-RNN-GAN for time series about keras-gan HOT 1 OPEN

This seems pretty hard to implement in keras as has a GAN, which is already overly complicated to implement in keras(as compared to tensorflow), and it also includes things such as pretraining, which again, are much easier to implement in tensorflow. However, some things, such as the time distributed dense layers and bidirectional lstms are much easier to implement in keras. Only if there were a way to have the ease of keras layers and the flexibility of tensorflow at once... Oh yes there is - tf.keras. You can use tf.keras in order to make this much more easily. However, tf.keras doesn't work with tensorflow variable scopes, and I am in the process of finding a fix for that.

from keras-gan.