pytorch_geometric
pytorch_geometric copied to clipboard
pyg-team
Sparse version of `nn.dense.dense_mincut_pool`
🚀 The feature, motivation and pitch
The nn.dense.dense_mincut_pool operator requires the input adjacency matrix to be dense. This requirement does not scale well as the matrix size grows. We may use operations on sparse matrix to address this issue.
Alternatives
Additional context
No response
If you see any major flaw in this idea, please let me know. I'd be glad to contribute to this idea.
Similar issue: https://github.com/pyg-team/pytorch_geometric/issues/7364
We initially designed it this way because mincut_pool needs to operate on dense adjacency matrices anyway in consecutive applications. I am definitely open though to support it for both sparse and dense adjacency matrices.
Hi, I'm interested in making a pull request. Meanwhile I cannot push my branch to this project. Maybe I need to be added to this project first?
If you want to contribute to PyG, you need to fork this repo, push the changes to your local fork, and then submit a PR to PyG.
Thanks for the tips. I made a draft PR https://github.com/pyg-team/pytorch_geometric/pull/9243 but have some questions. It would be great if someone could take a look at it.