Rex Ying

Hi, In terms of computation power, only pooled layers have fully connected structure, but the input level gnn does not. We tried to address your concern by calling this soft...

Duplicate of #6 Does this answer your question? The maximum cluster number is fixed. Let me know if there's more question

Hi, Guadalupe is right that it was cross entropy loss for link pred loss. The reason is that since the assignment prediction contains values between [0,1], cross entropy is more...

C is chosen to be, e.g. assign-ratio * num nodes, where assign-ratio could be 0.1, 0.25 etc. In practice, since graph sizes vary, it might not be the best number...

Hi, by default it's the number of nodes in the largest graph in the dataset. This is just for efficiency concerns. E.g. when you have a very large graph outlier.

I think you do not need to do aggregation of node embeddings into graph embeddings if you want to classify nodes. You can simply use GCN/GraphSAGE/GAT etc.

The two datasets have different labels.

Yes 5 classes for the 5K dataset; 11 classes for the 12K dataset. The original datasets are proposed by the Deep Graph Kernel Paper.

Hi, Thanks for pointing out. I think batch norm is quite confusing for GNNs and what you said make sense. I pushed the new version with bn having trainable parameters...

It won't work out-of-the-box. If we do not know the connectivity of new nodes, the network won't know how to propagate information from it's neighborhood. But I think if in...