Negative and Positive sampling with NeighborLoader

[k-s-b]

Hi @rusty1s ,

How to sample positive (one hop away) and negative nodes with NeighborLoader?

I want to perform unsupervised training by keeping the embeddings of positive nodes closer to the center nodes, and push embeddings of negative samples away. In the previous versions, it was possible to do it via NeighborSampler, as here.
I see that it is possible to perform negative sampling with LinkNeighborLoader but it samples negative edges rather than nodes.

[rusty1s]

I think this is exactly what our graph_sage_unsup.py example is doing, so please take a look. I don't think you can sample negative nodes without sampling negative edges?