RCWGAN

A CWGAN model with regression.
we proposed a novel virtual sample generation method embedding a deep neural network as a regressor into conditional Wasserstein generative adversarial networks with gradient penalty (rCWGAN). In rCWGAN, conditional variables are introduced making the training supervised and a dual training algorithm is specially designed. With the advanced structure and training algorithm, the model has powerful sample generation capabilities and can handle prediction problems of quality variable.

Environment

Python 3.6

Dependencies

  • numpy
  • pandas
  • matplotlib
  • tensorflow
  • keras

all packages need to be installed on a conda environment with python >= 3.0

Acknowledgement

We appreciate efforts in https://github.com/eriklindernoren/Keras-GAN and in https://github.com/mkirchmeyer/ganRegression.

GitHub

View Github