I am trying to train a predictive model on EEG signals, because of the high frequency of signals I am using a wavenet, after training when I use it as a generative model (like PixelCNN) it fails miserably and generates a bias only; it's 1.probably because of my loss function which is MSE and the network only learned to repeat the last input or 2.the network is not powerful enough to model my data.
so how can I tackle my problem?, currently I don't have a lot of GPU power and can not increase my model complexity, but with that said, are there any rule of thumbs for Wavenet parameters?
are there any better loss functions for this kind of problems? (I am going to quantize my input and output so simply copying the last input wont do the job but I don't think that it's enough) 
