Isn't there data leakage in this tutorial

Questions
#1

I was just going through this Stochastic Training of GNN for Link Prediction — DGL 0.8.2post1 documentation

I can see that the train_loader uses the whole graph and all the edges in the dataloader for the model to be trained.

Doesn’t that mean that the model is also seeing the validation and test edges? Since these edges are not being removed?

Could someone answer these questions please?
#2

Hi @ecoy. You are right. This is the transductive learning for link prediction.

#3

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.