AttentiveFP supernode missing

[alexbui91]

Hello, thank you for your work.

I have interest in the AttentiveFP implementation from the paper "Pushing the Boundaries of Molecular Representation for Drug Discovery with the Graph Attention Mechanism". The Google search result leads me to this github repo. According to the paper, after the first phase of GNN message passing, the model adds a super node to the graph and considers this derived version as a new graph. However, according to your implementation, the same graph is used for both two phases. Please correct me if I am wrong.

f.y.i: the pytorch geometric implementation seems to be more accurate with the supernode implementation:
https://pytorch-geometric.readthedocs.io/en/latest/_modules/torch_geometric/nn/models/attentive_fp.html#AttentiveFP

Thank you for your consideration!

[mufeili]

Thank you for the report. Unfortunately, I've left AWS and cannot update the codebase or approve PR from others. You may modify your own fork if you need to use this functionality.