xiaojiew94 / kdgan Goto Github PK

In this codes, there are no exact explanations for exact training setup.
And there is no cifar-10 evaluation code. Besides these codes are somewhat spaghetti.
Furthermore, according to these codes in utils and lenet_utils, the proposed models cannot be reproduced exactly. More than anything Horrible, Horrible readability.

Hi, I'm reading your code but I find there is no code about KDGAN on CIFAR10. Only MNIST is given. If possible, could you please give the kdgan code on CIFAR10 ?

What TensorfFlow version is needed to run the MNIST example? Thank you!

Traceback (most recent call last):
File "pretrain_gen.py", line 28, in
tn_gen = GEN(flags, mnist.train, is_training=True)
File "/Users/florinmatei/Projects/KDGAN2/KDGAN/mdlcompr/gen_model.py", line 35, in init
self.gumbel_labels = utils.gumbel_softmax(self.logits, self.gumbel_temperature)
AttributeError: 'module' object has no attribute 'gumbel_softmax'

There is a gumbel_softmax.py in mdlcompr folder but if I'm trying to use it directly (by importing it in get_model.py module) I'm running into issues with TensorFlow saying that tf.shape is not a valid method so knowing what TensorFlow version is another important thing we need to know.