Train GAT in an unsupervised way?

[ZRH0308]

Hi!
I need to train GAT in an unsupervised way like GraphSAGE https://github.com/pyg-team/pytorch_geometric/blob/b57489fd759273a4a7ba4a9613fa87831daa937d/examples/graph_sage_unsup.py,
is there any problem if I do this?

[ZRH0308]

I have tried this and got better results on Cora than GraphSAGE (acc 79% vs. 75%), but I'm not sure if there are any potential risks?

[EdisonLeeeee]

Look correct to me. Both GraphSAGE and GAT are inductive methods and they should be equivalent in most cases.

[ZRH0308]

Look correct to me. Both GraphSAGE and GAT are inductive methods and they should be equivalent in most cases.

Ok! Thank you for your reply!