Improve the doc about pykeen HOT 3 CLOSED

Hi @robertanto, thanks for being patient. We've just put out the new PyKEEN v1.0 today, with a complete re-write of everything. There's much more documentation (though it still has a long way to go since there are so many features now)

You can check out the pykeen.pipeline tutorial which will get you training your first model in a few lines of code, then you can start reading the pykeen.pipeline reference to learn more, and lead yourself to all of the tricky configurations possible for each model, loss function, regularizer etc.

At this point, your feedback will be valued greatly, as we're moving towards doing a documentation sprint.

I'm not sure about Graph Neural Networks... @sharifza has implemented R-GCN for PyKEEN 1.0 so you can direct any specific questions about that to him. Since PyKEEN isn't really a framework for building new tools like PyTorch or PyTorch Geometric, there might be a misunderstanding. You can always use those tools to implement models then make them fit the interface for PyKEEN (in the mean time, PyTorch lightning was published which does something very similar to PyKEEN)

from pykeen.

Hi @robertanto, I think you have mistaken Knowledge Graph embedding models with Graph Conv Neural Networks. Although as @cthoyt mentioned, R-GCN is based on Graph Conv Neural Networks, I'm not sure if your question is directed specifically to that model. If my assumption is right, I would direct you to this and this paper.

from pykeen.

Thank you both @sharifza and @cthoyt !

Yes, for me the R-GCN model is a sufficient example.

from pykeen.