Add [VNode]
Opened this issue ยท 4 comments
leffff commented
Add [VNode]
luke-a-thompson commented
Hi leffff, I'm wondering if you have implemented the VNode readout?
It's my understanding that the VNode is what enables graph classification tasks. Without the VNode is whole-graph classification still possible with your model?
Thanks
leffff commented
Hi!
Well, unfortunately I haven't :(
However you can still use any invariant aggregation function, like mean/max/min to pool your node embeddings into a graph embedding!
I plan on adding it!
luke-a-thompson commented
Thanks for the quick response leffff, I'll try mean pooling for my project :).
Good luck with adding the VNode, I know in the paper they say it can produce a significant performance uplift.
Thanks
StefanIsSmart commented
Looking forward to the updating of VNode.