Edge Features and Node Features

I’m a newbie here, so please pardon the simple question.

So most of the DGL code I’ve seen uses node features and targets, not edge features and targets. Is there a way to use both at the same time? How does this interact with graph convolution, etc? Typically, for example, we’d want to have the node messages paired with the edge state/features.

1 Like

Actually, in the application of molecule tasks, it is very common to use both the node features and edge features. You can have a look at the details of the models on those tasks.

1 Like

Yeah, I’m seeing some of it now…

Yes, there are many message functions that use both node/edge features. If you have some particular ideas you can elaborate them a bit more.

1 Like

I see now how edge features fit in. It just comes down to how precisely I’ll be able to encode the function I need. I’ll look in to it and see.