How to integrate own negative edges in TGN example?

[ArchitNK]

Hello, Thank you for the great library and support for Graphs.
I am working on a dataset that requires temporal context. So I am using TGN example from the PyG library.

1. I have a dataset where I have some interactions where there is no edge (Negative edges).
   I want to integrate them into training instead of the artificial Negative edge sampling from dataloader.
   In short I want my to define my own neg_dst. Is there any way to do that?

2. The current neg edge sampling generates random sample nodes where min(original nodes)< node < max(original node). This could create nodes that are not there at all in the original nodes.
   Eg. Original nodes[1,2,5] -> negative sampling could give [1,2,3,4,5]
   Is there any way to ensure that with negative edge sampling only nodes from original list would be created?

[rusty1s]

You can consider overwriting TemporalDataLoader to implement your own negative sampling strategy. Would that work for you?

[ArchitNK]

Yes, That could work. Just wanted to know if there is already a method that can do this within PyG

[rusty1s]

Currently no, this would be the recommended way for now.

[ArchitNK]

Thank you for the suggestion.