How to correlate node representation and graph representation?

[Bowen-n]

I have a GNN model with a global pooling method, so I can compute the node representations and the graph representations.
Now, for each graph in a batch, I want to update the node repr by correlating node repr and graph repr.

How can I do this for a batch of graphs?

[rusty1s]

You can consider concatenating both as part of your model:

x_global = global_mean_pool(x, batch)
x = torch.cat([x, x_global[batch]], dim=-1)

[Bowen-n]

It solves my problem, thanks !!