When should I stop a training of WGAN model?
The loss function of the WGAN is a continuous one. It doesn't have a convergence point. I don't really understand when we should stop the training.
See also questions close to this topic

How to get/evaluate weights from deep learning model
I would like to know is there a way to get/evaluate attributes' weights after training the deep learning model so that I am able to know which attribute may be a more important role to the question. I would be grateful for any help!

How to penalise my agent for using different action?
I'm using A3C in tensroflow, I want to keep my agent using the same action and only change when necessary and will have a benefit from changing.

what functions of OPENCV mainly used in image analysis or particularly in deep learning?
I want to know about what functions or how we use OPENCV in analysis the image which is overall beneficial for deep learning models?

How to generate time series data using GAN?
How to generate time series data (network monitoring) data using GAN? Is GAN is good for time series text data generation ?
For detecting the anomaly in network using log data we need to generate the text data.

Sampling From A Distribution for Training Adversarial Autoencoder in Tensorflow
Here is my understanding of the enforcement process of a distribution on the latent representation of an Adversarial Autoencoder:
To train an Adversarial Autoencoder, a sample is drawn from a prior distribution (like a Standard Normal distribution), say
p(z)
and the discriminator network compares this sample with the latent representationz
, modelled by distributionq(z)
(actuallyq(zx)
) to output something that tells ifz
is the 'real' or 'fake' sample from the prior distribution.How exactly is this part of the training process implemented in Python (Tensorflow)?
I understand that the discriminator network is built in the same way as any other network and is really specific to the type of data the latent representation holds. What I am more interested in is knowing what the inputs of the discriminator network are and how the samples are generated and fed to the network. Can someone please help with this?

What is the main idea of global conditioning in WaveNet?
Now, I want to use global conditioning to implement multispeakers. But I am confused about what is the main idea of global conditioning? Let me show you how I think of it and please point out where I am wrong or give me some suggestions or guidance. For instance, I have a data set containing 100 training samples and the length of each training sample is 2000. This data set has 4 types of samples and the labels are 0, 1, 2, 3. Now, I want to form an input vector with labels. After onehot coding, the size of data set is 100x256x2000 (mu=256). Then after onehot coding, the size of label of total training samples is 100x256x1. Finally, the input vector I want is is 100x256x2001, which means I should add the label vector in the form of onehot to the end of the data set vector in the form of onehot.
Am I right?