I have a list of graphs on which I want to perform node classification task. Each graph has nodes, node attributes, edges and associated labels for each node. Any guidelines on how to train this dataset?
Opened this issue · 1 comments
clydebailey commented
I have a list of graphs on which I want to perform node classification task. Each graph has nodes, node attributes, edges and associated labels for each node. Any guidelines on how to train this dataset?
qitianwu commented
Hi, we release an extended version of our follow-up work DIFFormer to accommodate a batch of graphs as input. The implementation allows for graphs with different sizes that are disconnected (the global attentions are computed within nodes in each graph). See the model implementation here https://github.com/qitianwu/DIFFormer/blob/extension/physical%20particle/difformer-v2.py
The extension for NodeFormer can be done in a similar way as DIFFormer.