CLN HDBSCAN _tree.pyx::do_labelling refactor
#26101
Conversation
Have you measured any impact on performance? |
There was a problem hiding this comment.
I think the do_labelling function is complex enough to be turned into a cpdef function to be testable with a few dedicated unit tests, at least for the nominal cases, on simplistic data and ideally on a few edge case.
Testing the nominal cases would serve as:
- documenting how this function is expected to work for future maintainers,
- and for current day HDBSCAN reviewers,
- non regression tests for future refactorings.
I also thinkg that a docstring would help (in addition).
Similar comment for the other functions / classes of this Cython module although probably in decidated PRs to keep this one focused.
|
@ogrisel I've added docstring and tests for |
|
@scikit-learn/core-devs If anyone has some extra bandwidth I would greatly appreciate a second review on this PR :) |
Reference Issues/PRs
Addresses #24686
Selected subset of #26011
What does this implement/fix? Explain your changes.
Changes variable names to new standard, and includes an algorithm refactor to
do_labelling. The new function is logically equivalent to the old, just with if-statement de-nesting and improved naming of intermediate values for readability.Any other comments?
These changes were extracted from #26011 to facilitate quick review