Generative Modelling on Graphs Request.

[joeybose]

I love Pytorch Geometric and it's ease of use over other graph libs. However, one pain point is that there doesn't seem to be easy to use graph generative models. In particular, it would be really to have some canonical Generative models on graphs which already have pytorch code. Personally, I would love if these ones could be added:

1) GRAN: https://github.com/lrjconan/GRAN 2) GraphRNN: https://github.com/JiaxuanYou/graph-generation

I'm happy to help build this together as well :)

[rusty1s]

I really like that idea, but I sadly do not have much expertise in graph generation. Eventually, @JiaxuanYou has some thoughts on how to integrate those models?

[suiluj]

@joeybose did you manage to use pytorch geometric for generative graph models?

[joeybose]

I did try it on a project but I quickly abandoned it as Graph Generation is quite a challenging task and the GRAN repo is already super opaque.

[suiluj]

ah okay thanks for the info.

[nashid]

@rusty1s @JiaxuanYou @joeybose wondering any of you built a generative model with this library?

[orausch]

@nashid I'm working on it (for another project), will try to open a PR later this month :)

[ketyi]

@nashid I'm working on it (for another project), will try to open a PR later this month :)

Keep me in the loop please :)

[nashid]

@orausch curious to know which generative model are you implementing?

[orausch]

@orausch curious to know which generative model are you implementing?

We have implemented the GraphRNN-S model from the GraphRNN paper