Percentile Pooling for Global Output

[khainguyen20]

I have a quick question. For the global pooling operations, I believe there are only 3 for now, which is global max, global average, and global add. Is there anything where instead, we can get the percentiles of the node features (25%, 50%, etc.)? I believe this way, we lose less information and is also less prone to outliers. I'm trying to implement this but I'm having a bit difficulty with getting the percentiles for the specific graphs in a batch. Would there be a way to implement this type of global pooling operation? Please let me know. Any help would be appreciated!

[rusty1s]

There exists a QuantileAggregation operator you can use, see here.

[khainguyen20]

This is what I was looking for. Thank you!