[FEA] UMAP umap_learn integration metaissue

[lmeyerov]

Is your feature request related to a problem? Please describe.

Using umap_learn with graphistry should be more visible and easier!

See: https://github.com/graphistry/pygraphistry/blob/master/demos/demos_databases_apis/umap_learn/umap_learn.ipynb

Describe the solution you'd like

• [ ] umap_learn: embedding.plot() <- maintainers already signaled interest
  • [ ] method
  • [ ] docs on their side
• [x] pygraphistry: g = g.nodes(...).umap(engine='umap_learn', params=...); print(g._embeddings); g.plot()
  • Unclear what input params + outputs for typical cases... maybe just raw parent objs until we understand what to standardize?
• [ ] attributes: options not just for embedding, but original properties + potentially vectorized features
• [x] option to autovectorize, nodes & maybe label propagated edges <-- feels like a separate issue
• [ ] maybe some recommended settings or enrichments, like:
  • compact layout
  • color edges by similarity

Implementation notes

May be easiest if mostly implemented as a shim to our side, so we can update with less friction, and they just version bump whenever ready

Try to keep varargs in most places to be more robust against changes on either side

[lmeyerov]

@lmcinnes FYI :)

[lmeyerov]

Moving discussion to https://github.com/graphistry/pygraphistry/discussions/300