How to check gradients for ensemble-like architecture?

autograd
1

Hello,

I am making an architecture with 2 Autoencoders. The input is passed to all auto encoders but only the autoencoder with the best (lowest) reconstruction is considered in the loss function. This should result in gradients update for that one autoencoder, no? How can I check the gradient flow in PyTorch for such a model? I have other things I want to check (e.g. are my gradients vanishing). Is PyTorch set up in a different way, such that I will need to use autograd to achieve this? ( I would rather not have to change anything with autograd)

I am new and have been looking around but have not found anything.
Thank you!

2

Hi, @fabrizio_chavez
Can you provide a code, pseudocode, or something similar to clarify what you actually need?

3

Hi, thank you for getting back at me! I’ve actually figure this out in the time between the view and your response. I will close this thread.

1 Like
4

Ok, No Problem. It’s Nice to see that you figure out your self.

1 Like