snap-stanford/neural-subgraph-learning-GNN

question on subgraph alignment code

shopeth opened this issue · 1 comments

Hello,

Thank you for sharing the code for the interesting work. I got a question on this part of the alignment code:

batch = utils.batch_nx_graphs([query, target], anchors=[u, v])

Is the query and target position switched? In the model forwarding, it seems the first position pos_a is the supergraph of second position pos_b. Current work will generate the opposite prediction.

qema commented

Hi yes thank you for the catch :) our internal code had this fixed, but I had forgotten to update the public repo. We are working on an improved version with more robust alignment code and support for node features. Updated the current repo to use the correct ordering of query and target