We've implemented a GCN which seems to be the first type of GNN.
Since then, there have been GraphSAGE and Graph Attention Network (GAT) architectures that build on top of it.
From a quick glance at them, it seems like GraphSAGE is more useful on very large graphs, so could be overkill for us, but GAT seems promising for our purposes.
We've implemented a GCN which seems to be the first type of GNN.
Since then, there have been GraphSAGE and Graph Attention Network (GAT) architectures that build on top of it.
From a quick glance at them, it seems like GraphSAGE is more useful on very large graphs, so could be overkill for us, but GAT seems promising for our purposes.
GraphSAGE: https://github.com/williamleif/GraphSAGE GATs: https://github.com/PetarV-/GAT