SeqGAN

This is the basic demo of the SEQUENCE GAN implemented by PyTorch for personal study.
The original versioin of the tenensorflow is in here

Requirements

PyTorch 0.4.0
Visdom 0.1.7

Run

Initialize the Visdom first.

$ python3 -m visdom.server

Then run by

$ python3 main

if using cuda then pass the cuda number.

$ python3 main --cuda=0

Then you can view the training curve in the visdom environment.

Overview Model

Learning Curve

Adversarial training after 40 epoch supervised pretrain of the generator.