giannisnik
commented
4 years ago Dear @sbhttchryy ,
All the kernels contained in GraKeL are kernels between graphs, i.e., each sample is a graph and not a node.
If in your problem, classes capture structural similarity (and not homophily), then you can follow this approach:
- create one graph for each node (e.g., its egonetwork of radius r)
- formulate the problem as a graph classification one (instead of node classification)
- use kernels from GraKeL along with SVM to perform classification
Hello, I have a Networkx multigraph that I have created using the AIFB dataset. I want to use the kernels to perform classification tasks on nodes. Since these graphs have edge labels, do you have any suggestions as to what kernel might be good for this case?