Could you kindly suggest how to extract edge weight from DGI model (dgl/examples/pytorch/dgi at master · dmlc/dgl · GitHub)? DGI has used mutual information to build the model, therefore, the weights in GCN (A part of ) might not equal to the weights of DGI.
I assume this is resolved via alternative channels. The answer is that you may use an attention-based model for encoder in DGI, in which case you can retrieve the learned attention. Feel free to reopen it for a follow-up discussion.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.