rev edge type confusion

[Klahadore]

Hello
if I use ToUndirected(), will I need to manually add the edge attributes for the reverse edge, or will ToUndirected carry them over.

Also for self attention, if is do something like Linear(torch.cat(x_i, x_i)), or do i define add self loops to the dataset, meaning i will also have to add edge attributes for self attention?
Thanks in advance

[rusty1s]

if I use ToUndirected(), will I need to manually add the edge attributes for the reverse edge, or will ToUndirected carry them over.

They will be carried over.

Also for self attention, if is do something like Linear(torch.cat(x_i, x_i)), or do i define add self loops to the dataset, meaning i will also have to add edge attributes for self attention?

Not totally sure I understand. If you define self-loops in your graph and your graph has edge features, then the self-loops should also define edge features.