After training a model on GCN (such as dgl/link_predict.py at master · dmlc/dgl · GitHub), how to extract edge attribute of the model? Which kind of model can be extracted numeric edge weight?
Thank you for your help!
Models like GCN do not explicitly yield edge weights. One possibility is to take the representations of the incident nodes of edges and compute a score from them, e.g. you can compute the inner product of a pair of node representations. Alternatively, you can train models that explicitly yield edge weights like GAT.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.