Conditional generation during sampling in binary class case

[PradKalkar]

Hi. I was wondering if its possible to do conditional generation during sampling, i.e. I am currently having a binary classification task and I would like to specify how many rows with label 1 I need to generate and similarly for other label. This time, I would be providing rows belonging to both the classes while training the GAN. Is it possible to specify such a distribution during sampling so that I get desired number of positives and negatives?

[zhao-zilong]

Hi @PradKalkar in our testing code, we actually did that, but we didn't provide this in our codebase, we don't want to make our code too complicated. It should be quite easy to achieve that, just made the conditional vector by yourself instead of random sampling. Check this line:

CTAB-GAN/model/synthesizer/ctabgan_synthesizer.py

Line 803 in 3e1d80c

condvec = self.cond_generator.sample(self.batch_size)