DeepGraphLearning
Is there or would it make sense to have more integration with Pytorch Geometric?
Hi! This is perhaps a silly question, but would it make sense to have integration with Pytorch Geometric: https://github.com/pyg-team/pytorch_geometric ? I think there could potentially be few advantages, such as easier maintenance, less duplicated work, and even easier usage with things such as GraphGym. But of course, there are also a few drawbacks such as less control; and would probably entail significant refactoring I believe. What do you think?
Hi! That's a great initiative. However, just as you noticed, it could take very signifcant refactoring and I guess the team is not ready for that. Also some of the designing goals in TorchDrug may violate the goals of PyTorch-Geometric. For example, TorchDrug is designed to be molecule-centric (and maybe protein-centric in the future). I am not sure if PyG will accept such a domain-specific design in their general graph library.
As a compromise, we are considering providing some functionalities so that users can easily import PyG models in our library. What do you think?
