Is is possible to use a vector as edge feature when using GAT for graph classification?

Questions
,
#1

Hi, I’ve used both node and edge features for GCN based graph classification before, and it worked well. But I’m wondering if I can use both of them in GAT models.

Attention scores can be treated as edge feature/weight, right? And I didn’t see the tutorial dealing with edge feature.

Is it possible? Is there some blogs or codes I can refer to? Thank you.

#2

Are these edges weights or non-weight edge features?

#4

Thank you for your reply. They are non-weight edge features.

#5

I think you can combine graph.edata['e'] with your edge features here.

#6

Thank you. I will try.

#7

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.