I have tried to solve the node classification problems by graphSAGE and it works well on dataset Cora. Now I want to convert it to solve the regression problem, and I only change the loss function into torch.nn.MSE() without any change to other parts of the code.
However, the loss does not decrease. So I'm wondering if graphSAGE can be used to solve regression problem and how should I convert it?
ps:
My code is from this website:https://github.com/ytchx1999/GraphSAGE-Cora-Citeseer-Pubmed/tree/main
And I only change the dataset and convert criterion = nn.CrossEntropyLoss().to(DEVICE) to criterion = nn.MSE().to(DEVICE).