Transductive Approaches

[BaharanKh]

Hi, thanks for your tutorial!

I am confused by something you mentioned in it. You said GCN is transductive, in contrast to GraphSage, which is inductive. Later you mentioned one of the shortcomings of the transductive methods as not being able to use node features. I was wondering if this statement is correct since GCN takes adjacency matrix along with feature matrix as its input and the features will be the hidden representations at level 0, right?

I would be grateful if you address my confustion. Thanks!