How to perform node classification using Graph Neural Networks

Question

I'm am trying to perform node classification using graph neural network methods. My initial plan was to convert my graphs to adjacency matrices and train my network on that, with the node features being my target.

However, my graphs all have a different number of nodes, so I believe adjacency matrices will not work. I then found information about node embeddings and applications in biology (see here). It infers here that embedding your nodes no longer matters about graph size.

What I am wondering, is if using embeddings is actually the solution to my problem? Then, how, practically, do I build a Graph Neural Network out of this method? The link (in the slides for part 2) still says I need to train my network to learn the embeddings - but now I have my original problem, how can I train the network on many different-sized graphs?

Doing node classification on small, differently sized graphs seems to be somewhat commonly done. What I am slightly confused by is the step from theory to practice.

Could anyone please offer any advice? Or links to more material that could solve some of my queries?

Answer 1

I'm not sure if that's what you need but this library seems pretty similar to your problem: https://github.com/Accenture/AmpliGraph/

Answer 2

you can use some thing like DeepWalk to embed each node so it is straightforward to do any classification on that embedding as it is just a vector.

Answer 3

Keras now provides better GNN intergation and now provides a relatively complete tutorial for this exact purpose: enter link description here.

If anything I have used this code and was able to calibrate a GNN that gave beter performance than the basic NN approach.