Refactor graph lifting - Githubissues

NSAPH-Projects / topological-equivariant-networks

E(n)-Equivariant Topological Neural Networks

MIT License

19 stars 0 forks source link

Refactor graph lifting #6

Closed mauriciogtec closed 7 months ago

mauriciogtec commented 8 months ago

Graph lifting in ESMPN needs refactoring (e.g., it has problems with memory management for the Vietoris complex).

TODO: detail the remaining issues.

mauriciogtec commented 8 months ago

Several enhancements were introduced in #5. @ekarais What is still missing for graph lifting?

ekarais commented 8 months ago

removing the implicit rank = cardinality constraint
memory issue (can be resolved by passing the graph lift transform to the dataloader in qm9.utils.generate_loaders_qm9 as opposed to the QM9 dataset), this should not be too complex

These are refactorings specific to graph lifting. Another important refactoring will be how we handle cell features. i.e. heterogeneous feature spaces for cells of same rank.

ekarais commented 7 months ago

I will close this issue and instead create a specific issue for the memory problem.