self.log_scale in line 34 in vae.py should be a vector of size latent_dim

[shubhamguptaiitd]

Hello William, This is a very good code, the only issue is that initially, we were not able to train on our dataset but after reading the VAE paper, we found that self.log_scale parameter should be a vector of the size of output dimension instead of a real no. . After correcting this our code started to train. Please see if this finding is relevant here.