wgan-gp getting different rmse and prediction not accurate

[makovez]

I followed (copy/paste) the exact same code of wgan-gp_demo.ipynb with the same dataset as well, but I am getting worst result. Are you sure the code is updated? I'm getting RMSE:7.945424793070616 on testing dataset

Also, As you can see the trend of the prediction/actual seems much more similar to GAN or LSTM

[makovez]

Running it multiple times changes the score. Is this normal?
It doesen't seem to be much stable. RMSE in testing ranges from 2.5 to 10

[ChickenBenny]

Hi makovez,
Because GAN in a non-steady model, so the RMSE and the result will be different.
In other words, GAN is depend on the interaction between G and D, just like reinforcement learning.
Therefore, the result might be different.
If you are interesting in this topic try to use MHGAN or use Bayesian optimization, it might help to improve the training.

[makovez]

Hi, it would be nice if we could have a chat regarding the topic? In particular I would like to exchange some ideas about feature selection and labels generation with RL.

My telegram is: @sbongown

[ChickenBenny]

Hi, just feel free to ask me~
I have already added your telegram.