Interesting. I am not an expert in those algorithms -- though I have used e.g. node2vec. The large number of approaches and subtle difference suggest both that embeddings are interesting to lots of people and that they are solving many hard problems.

We do have some embedding tools, but they mostly focus on 2D embeddings (drawing layouts). And we have some tools for higher dimensional embeddings, but nothing like the focus on high-dimensional embeddings that comes from the machine learning applications. Is that where these algorithms come from?

I have a healthy respect for the mathematics behind the problems of approximating probabilities and generating random walks which these seem to rely on. It may be that getting them to work well is difficult enough to justify them being in separate libraries from networkx. That said, I am guessing a library that focused on embedding algorithms would be quite popular if only because people would have a go-to-library rather than having to try each of the tools out separately. And yes, it could be built on top of networkx (with probably many other tools from the Scientific Python toolset). A quick search shows a few existing discussions and repos that implement node2vec acting on a networkx graph. It might be worth spending some time looking through the current ecosystem to see where your efforts can give the biggest payoff. Building a library that brings together those algorithms may be something that others are willing to help out with.

I can also imagine that some tooling in networkx would help those efforts. If you find something like that, we are interested in adding features to NetworkX which make constructing and combining these algorithms easier. But it feels like this is better as an upstream project rather than being housed in networkx.