Revise GNNLux Temporal Graph Layers

[juanmigutierrez]

Hi @CarloLucibello,

I’m excited to contribute to this project, especially as I’m interested in working on it as part of Google Summer of Code.

I am currently pursuing a Master’s in Mathematical Engineering at Politecnico di Milano, and contributing to a library focused on Graph Neural Networks (GNNs) aligns well with my academic and professional interests. This experience will allow me to deepen my understanding of state-of-the-art GNN models while also helping improve the community’s access to this library.

My background includes two Bachelor's degrees in Applied Mathematics & Computer Science and Economics. I have 4 years working in Industry as Data Scientist (Now only focusing in my studies) .and also spend my free time researching about differential privacy with some professor on the University de los Andes (Colombia), we hope to publish soon a new algorithm we developed. I think you can check my linkedin for further review of my profile: https://www.linkedin.com/in/juanmigut/

Over the weekend, I started diving into this topic, has been very cool so far! and believe I can contribute in the summer by improving tutorials,addressing issues related to CUDA and sparse linear matrices (I have prior experience with CUDA, get 2nd place in a Hackaton for Oracle on a CUDA problem) or any other cool idea I might get for the proposal. While in the meantime I plan to help solve issues to understand more the library in its deeps.

Changes in this commit:

• Fixed the Flux implementation in the tutorial case
  • I encountered some errors and resolved them. Open to discussion if any refinements are needed.
• Implemented a Temporal Graph Neural Network tutorial following the GNNLux framework
  • This is an initial draft, and I plan to review it further in the following days.
  • Currently, the training loss improves over epochs, but the test performance does not.

Issue reference:

🔗 Resolves issue #562

I would appreciate any guidance on the next steps to contribute more effectively.

Looking forward to your feedback! 🚀

[CarloLucibello]

I suggest to open separate PRs for 1) flux tutorials 2) revising lux temporal layers 3) lux tutorials

[juanmigutierrez]

I think I needs more instructions to fully solve the Temporal Graph Layers problem.

I'll focus on the non-linearity of TGCN issue #591 , which seems simpler and clear to me. So will close this pull request and create the other one, to leave free this one for another person if can do it before.

I think I will have this things flux and tutorials, maybe I can put it on the proposal to, to make a full proyect hard in the GSOC, for the meantime will upload the other pull request with the feature.