Neighbour Samplig for edge classification

Questions
#1

Hi!

I’m working on an edge classification task, Ive already seen the documentation example, but im still not sure, what is the correct way to split a dataset that consists of a single graph for edge classification? I want to use neighbor sampling during training. How can I split the graph to separate the edges into training, validation, and test sets while applying neighbor sampling specifically during training?

#2

you need to split the dataset(namely edge ID or node pairs) into train/val/test on your own before instantiating dataloader.

#3

What I ended up doing was creating a train mask and a test mask for the edata, then extracting their IDs and passing them to the dataloader with neighbor sampling.

Thank you for helping me understand!