Fea/umap refine

[AjayThorve]

This PR aims to add UMAP refine function, to facilitate iterative fitting without using callbacks, which makes it possible to run UMAP in the node-rapids bindings and show iterative training visualizations. Related node-cuml PR here.

• adds UMAP refine & get_graph (to only calculate the cgraph computed in UMAP::fit once).

cc @trxcllnt , @exactlyallan

[cjnolet]

I like this design because it allows us to compute the connectivities graph individually from the embedding. We've been also talking about exposing a function like get_graph through the Python layer because, for example, other libraries such as Scanpy call fuzzy_simplicial_set on the UMAP estimator directly.

Overall, the changes are great and very straightforward. Just a couple very minor things.

[cjnolet]

LGTM, pending CI

[trxcllnt]

rerun tests

[trxcllnt]

rerun tests

[AjayThorve]

rerun tests

[dantegd]

@AjayThorve @trxcllnt PR #4217 needs to get merged for CI to pass, so I canceled the job, will re-trigger once that's in

[AjayThorve]

@AjayThorve @trxcllnt PR #4217 needs to get merged for CI to pass, so I canceled the job, will re-trigger once that's in

Sounds good, thanks!

[dantegd]

rerun tests

[dantegd]

@gpucibot merge

[AjayThorve]

rerun tests

[codecov-commenter]

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.10@9f9e16f). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-21.10    #4228   +/-   ##
===============================================
  Coverage                ?   86.19%           
===============================================
  Files                   ?      231           
  Lines                   ?    18837           
  Branches                ?        0           
===============================================
  Hits                    ?    16236           
  Misses                  ?     2601           
  Partials                ?        0           

Flag	Coverage Δ
dask	47.06% <0.00%> (?)
non-dask	78.94% <0.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 9f9e16f...44a9f39. Read the comment docs.