Pytorch_geometric: How to use global_max_pool with one graph?

[jiaruHithub]

❓ Questions & Help

I only have one graph which is have a feature matrix with n-cols and d-rows in one batch so that I don't know how should I set parameter 'batch' in this case?

global_max_pool(x, batch, size=None)
I don't know how to set the 'batch' in this case

Looking forward to your reply,thank you very mach!

[rusty1s]

x.max(dim=0) or set batch via batch = x.new_zeros(x.size(0), dtype=torch.long)

[jiaruHithub]

x.max(dim=0) or set batch via batch = x.new_zeros(x.size(0), dtype=torch.long)

Impressive!
Thank you for such a quick respond!

I have anthor question.
In every batch,I have N graphs that means I have a (N * n) * d matrix(every graph have n nodes,but n is different between every graph).I want make the dimension of every graph form n * d become
(n * d) * 1,
final,the matrix of batch(N graphs)'s dimension become N * (n * d).

I want use torch.bincount and torch.split the matrix,but I think the method is too slow.

Do have anthor methods?

Looking forward to your reply,thank you very mach!

[rusty1s]

That's what torch_geometric.utils.to_dense_batch is doing :)

[jiaruHithub]

That's what torch_geometric.utils.to_dense_batch is doing :)

OK, I'll try it later.
Finall, I want to ask why the parameter 'batch' can set like you told me ? I don't know the reason.

[rusty1s]

For each node, batch holds the index of the graph the node belongs to. Since you only have a single graph, setting batch to zero assigns all nodes to the same graph (with index 0).

[jiaruHithub]

For each node, batch holds the index of the graph the node belongs to. Since you only have a single graph, setting batch to zero assigns all nodes to the same graph (with index 0).

I'm flattered that you're reply so quickly.
Thank you and I wish you a happy life