This repo contains pytorch implementations of several types of GANs, including DCGAN, WGAN and WGAN-GP, for 1-D signal. It was used to generate fake data of Raman spectra, which are typically used in Chemometrics as the fingerprints of materials.
If you use these codes, please kindly cite the this repository.
- python 3.7.8
- pytorch 1.6.0
- numpy 1.19.2
- matplotlib 3.3.0
NOTE: RMSprop was used in the implementation of wgan-gp, rather than Adam, which was used in its original version, as it seemed like Adam didn't work well in my applications.
You need to put all your data in the same folder in txt format, and make sure they are column vector.
The length of signal in my application is 1824, you need to modify the networks according to your data.
Run python file end with "train".
[1] Nathan Inkawhich. DCGAN Tutorial [EB/OL]. https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html, 2020-10-14.
[2] Yangyangji. GAN-Tutorial [DB/OL]. https://github.com/Yangyangii/GAN-Tutorial, 2020-10-15.
[3] mcclow12. wgan-gp-pytorch [DB/OL]. https://github.com/mcclow12/wgan-gp-pytorch, 2020-10-15.