Activation function in GCNConv layer.

[errhernandez]

Just to check that I understand: in the Kipf and Welling paper (https://arxiv.org/abs/1609.02907) describing the GCNConv, I understand that there is an activation function applied, like here (the sigma)

But in the PyG implementation the activation function is missing; I presume this is because one can just apply it to the output of the GCNConv layer, right?

Sorry if this is a stupid question. I'm a slow learner ;-)

[rusty1s]

Yes, activation functions are not part of our nn.conv layers. This has mostly two reasons:

• It makes it easier to try out different non-linearities
• It aligns well with operators from PyTorch, e.g., Conv2d, which also do not apply a non-linearity