Global Pooling

[hadarshavit]

Hi, I'm using GNNs for graph classification. Therefore, I need to aggregate the node embeddings. In PyG there are the "basic" pooling layers: max, mean, add.
I'm looking for more advanced pooling layers such as global attention pooling (like in https://arxiv.org/abs/1511.05493), but could not find an implementation of it.
Is there an implementation of more advanced pooling layers in PyG?
Thanks!

[EdisonLeeeee]

You can take a look at here, where some advanced pooling layers were introduced.

[rusty1s]

Please also take a look at this. All aggregation modules can also be used for global pooling.

[hadarshavit]

@rusty1s Thanks, is there a tutorial on how to use the aggregations as global pooling?

[rusty1s]

I will add this piece of information to the documentation. In general, it is as easy as doing:

aggr = MeanAggregation(...)

aggr(x, batch)