Example section for graph based generative models

[ritvikvasan]

🚀 Feature

I was wondering if there is a plan to provide examples for graph based generative models/if you think that the field is not mature enough when compared with node representation learning/link prediction. For example, I see that graph autoencoders with the inner product decoder are provided as an example of the link prediction task. It would be nice to see an example of the same graph autoencoder but with a probabilistic decoder to show capabilities in graph generation.

Let me know what you think!

[rusty1s]

I'm absolutely open to adding graph generation examples to PyG. Regarding your suggestion, is there a paper that already implements your idea? Are you interested in contributing such an example?

[ritvikvasan]

Thanks for the feedback! Yes, some papers that I have come across on this topic are the GraphVAE (https://arxiv.org/pdf/1802.03480.pdf) and GraphRNN (https://arxiv.org/abs/1802.08773). I would love to work on formulating this code using pytorch geometric..

[rusty1s]

Thanks for sharing. GraphVAE should be straightforward to add besides the internal assignment/graph matching step via integer quadratic programming.

[ritvikvasan]

Yes! Thanks. I can work on adding this as a PyG model.

[nashid]

@ritvikvasan has it been made into the library?

[ritvikvasan]

Unfortunately no! Sorry about that...