How to: Fully supervised learning on a single graph?

[a-0-ls]

Hi,

A very general question. If I have a single graph, is it possible to do supervised learning instead of using masks to do semi- supervised learning?

In this case, how can I make my test dataset?
I have networkx data object as my input graph. How can I get the test set out of this?

I have seen the randomlinksplit and randomnodesplit, my use case is for solving a node regression problem.
Any feedback would be very helpful.

[wsad1]

First you'll have to convert your graph to a torch_geometric.data instance using from_networkx.

Then, you could use subgraph to split your graph into separate train and test graphs (subgraphs).