Concat graph embedding to all node embedding before prediction

[jomalla123]

Hi,

Using some GNN layers I have embeddings for each node $h_i$. I would like to concat the graph_embedding $G$ (e.g. mean of all $h_i$) to each $h_i$, before passing this to some linear layers to make a prediction. However, since I working with a batch of graphes with different sizes, I have a hard time figuring out how do that. Is that even possible?

[rusty1s]

You can do this via

h = torch.cat([h, h_global[batch]], dim=-1)

[jomalla123]

Thanks for the reply, that it very convenient