search

PetarV- / GAT

Graph Attention Networks (https://arxiv.org/abs/1710.10903)
https://petar-v.com/GAT/
MIT License
3.18k stars 642 forks source link

About data #13

Closed SongFGH closed 5 years ago

SongFGH commented 5 years ago

Hi, I know the GAT model can apply to node classification. And in your paper, the node classification is executed in transductive learning. However, why your data is ind.cora.xxx? This data is used in inductive learning!

PetarV- commented 5 years ago

Hello,

If you observe how the data is constructed: https://github.com/PetarV-/GAT/blob/master/utils/process.py#L45-L96 all features end up being visible to the training algorithm, so it is a transductive setup. The inductive details are there solely in order to compute the necessary label masks.

It is, in fact, constructed identically as in Thomas Kipf's GCN repository.

Thanks, Petar