Which GNNs in dgl are good to try for graph classification?

Hello guys

so far i have used GCN, GIN, GAT, GraphSage for a graph classification task, i want to see which of the GNNs perform best on my dataset

so if you had to suggest 2-3 other GNNs for me to try out in a graph classification task ( for example by averaging the node representations and giving it to a linear classifier), which of them should i try next?

there are a lot of different methods implemented in https://github.com/dmlc/dgl/tree/master/examples/pytorch

so i am not sure which of them should i try next, i only want to try 2-3 more.

  1. I think you are already trying diffpool.
  2. DGL-LifeSci is a DGL-based package for life science. It implements models like MPNN, Weave, AttentiveFP for molecular property prediction, which is essentially graph regression/classification. These models require both node and edge features, but you may find some components worth trying.
  3. You can also find some models for graph classification here, though not necessarily based on DGL.
1 Like

Are edge features essential for DGL-LifeSci models? will they work properly if i don’t have edge features in my graph dataset?

It depends on the models. You may try removing them and see how that works.