MAINT: mutual information using upstream KDTree #31347
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
90d9ed9 to
d30b4fc
Compare
tylerjereddy
commented
May 10, 2025
| ny = kd.query_radius(y, radius, count_only=True, return_distance=False) | ||
| kd = KDTree(y) | ||
| ny = kd.query_ball_point(y, radius, p=np.inf) | ||
| ny = [len(sub_list) for sub_list in ny] |
There was a problem hiding this comment.
Here and elsewhere in this diff, we can do even better (see below). It seems that the two APIs have effectively developed via convergent evolution to offer similar options with different names.
kd = KDTree(y)
- ny = kd.query_ball_point(y, radius, p=np.inf)
- ny = [len(sub_list) for sub_list in ny]
+ ny = kd.query_ball_point(y, radius, p=np.inf, return_length=True)* This patch aims to test the waters for reducing community duplication of effort/maintenance with KDTree. In particular, a few select use cases of the `sklearn` in-house `KDTree` by the mutual information infrastructure are replaced with upstream `KDTree` from SciPy. The full test suite appears to pass locally.
d30b4fc to
da64dc8
Compare
|
I applied the simplification mentioned at #31347 (comment) and tests were still happy locally. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
KDTree. In particular, a few select use cases of thesklearnin-houseKDTreeby the mutual information infrastructure are replaced with upstreamKDTreefrom SciPy. The full test suite appears to pass locally.If there is an appetite for this, we could spend some time on it (scientific-python/summit-2025#31). I've placed some more detailed discussion points below. This may also interest @sturlamolden.
More detailed Analysis of the Situation
Potential Advantages of Doing This:
KDTreehas been heavily optimized and supports concurrency in many places that thesklearnversion does not, as far as I can tell.KDTreecould focus their efforts on a single (or at least, less) implementation(s). This is the main one I had in mind.Considerations, Drawbacks, Points that are not clear:
KDTreesources and offerings alone for quite some time to see how that goes first, progressively performing replacements in cases where there's an obvious benefit, or at least no loss of performance or functionality.KDTreecode is quite complex. The features offered by both libraries are not identical, and if you have a code base using two types ofKDTreeit may temporarily make things even more complex.asvbenchmarks for yourKDTreeimplementation I don't think? We'd want to add that to avoid performance regressions I suspect, and/or to demonstrate improvements where SciPy adds theworkers/concurrency option.sklearnuses a "base"BinaryTreethat is "specialized" toKDTreeandBallTreein an object-oriented style. SciPy doesn't haveBallTree(should it?)--how would this be coordinated if there was an appetite for upstreaming?sklearnactivity surrounding 32- and 64-bit "variations" or type preservation with the binary tree infrastructure? I'm not so sure how this would fly/work for SciPy.