Edge Classification without node features

[QianChang-page]

Thanks for your interest in my question.
I have a dataset including edge connection matrix (Data.edge_index) , edge features (data.edge_attr) and edge label (data.y).
I'm clueless on how to set up my model to achieve this Edge Classification.

[rusty1s]

In this case, you can either learn a shallow embedding matrix via torch.nn.Embedding to act as node features, or create a one-vector that acts as a dummy node feature.

[QianChang-page]

Thanks a lot, I treat the actual edges as “nodes” to be linked, and each “node” is the origin edge,the edge features matrix can act as node features .This turned out to be OK for me.

[rusty1s]

Yes, this works as well, in which case you are working on the line graph representation.