Jumping Knowledge Setting Last vs None

pyg-team / pytorch_geometric

Graph Neural Network Library for PyTorch

https://pyg.org

MIT License

21.31k stars 3.65k forks source link

Jumping Knowledge Setting Last vs None #8878

Open Bobowner opened 9 months ago

Bobowner commented 9 months ago

📚 Describe the documentation issue

It is unclear what the difference between None and last for the parameter jk is: https://pytorch-geometric.readthedocs.io/en/latest/generated/torch_geometric.nn.models.GCN.html

since the JK Paper only describes cat, lstm, and max.

Suggest a potential alternative/fix

No response

rusty1s commented 9 months ago

Sorry for the confusion. If jk Is specified, the model will additionally apply a final linear transformation to transform node embeddings to the expected output feature dimensionality. As such, jk="last" will perform a final linear transformation, while jk=None will not.