Question regarding Batchnorm about convlstmgru

[aserdega]

Hello.

So, first thing is that nn.init.xavier_uniform_(self.conv.weight, gain=nn.init.calculate_gain('tanh')) does initialization for parameters of convolutional module, not for batch norm.
Just take a more careful look at the parameters that are passed into initializer :)

And regarding batch norm, it has been a while since I've wrote this code. I don't see that bn modules are really used inside of RNN. I probably just forgot to delete this modules from cell class...

But if you gonna consider using batch norm, you want to apply it to the final output of the cell. In the very end of forward function.

Feel free to ask if there is anything else not clear to you. And if you find my code useful, I'd appreciate if you star the repo :))