I am following your tutorial on training GCN and I saw that you construct 1 graph object (called
g) with train/test/val masks and then
g is passed to GCN model. Do you have any example on training in inductive setting (like PPI) where the train and val/test graphs are separate? How can I pass 2 different graph object to the model? Or do I have to combine them into 1 graph object to be passed into GCN model? Thanks
See the answer to FAQ #15.
I saw on this example, all graphs are combined into 1 disjointed graphs and we use masks to distinguish between train, test, and val examples. Is there an example to train with multiple graphs (like in PPI dataset where we have 20 train graphs, 2 test graphs, and 2 val graphs? Or do we need to combine them into 1 disjointed graph?
Yes, we have a GAT example on PPI dataset here: dgl/examples/pytorch/gat at master · dmlc/dgl · GitHub
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.