Why adding bias at the output of every GCN Layer?

[utkarsh0902311047]

Hi,

1. Why add bias at the output of every GCN Layer?

I have one more doubt and want confirmation whether I am thinking correctly.
Does conv2 at the end of the model behave as a linear layer? Then cross_entropy will perform the log_softmax and Negative LogLikelihood Loss. Is this correct?

[rusty1s]

GCN would reduce to a Linear layer in case edge_index is empty, so it is naturally to also add a bias to it. You can customize this via bias=False when initializing the layer.