Creating test_graph in rgcn

Hi, I was wondering about this line of example code on link prediction using rgcn:

I’m not sure why the test_graph is built by passing the train data to build_test_graph()? Can someone clarify this please?

Hi @sharifza,

So the dataset is partitioned into training, validation, and testing. The training partition is something we already know before training, and the validate and test partition is what we want to predict. So here, when building the graph for testing / validation, we use all the knowledge we already have (i.e. the training data). Then on this full training graph, we try to predict the links in test partition.

2 Likes