FIX Allow for KMean's attributes to be readonly #24258
Conversation
ogrisel
left a comment
There was a problem hiding this comment.
LGTM, thanks for the fix.
Can you please just update the docstrings of lloyd_iter_chunked_dense and lloyd_iter_chunked_sparse to make it explicit that centers_new and centers_squared_norms can be left to None if update_centers is False?
ogrisel
left a comment
There was a problem hiding this comment.
I meant to give my approval instead of just a comment in the above review.
|
@jjerphan @jeremiedbb or @fcharras might be interested in giving this PR a second look. |
There was a problem hiding this comment.
LGTM. Thank you for fixing this issue.
Side note: I think the new test in this PR for read only attributes can become a common test.
I am OK making the test a common test. IMO, this can be done as part of this PR.
Side note: we recently discovered that some part of the routines can be simplified, see: #24262
Co-authored-by: Julien Jerphanion <[email protected]>
jeremiedbb
left a comment
There was a problem hiding this comment.
LGTM as well after this comment #24258 (review) is addressed
|
Updated PR with:
|
Co-authored-by: Julien Jerphanion <[email protected]>
Reference Issues/PRs
Fixes #24253
What does this implement/fix? Explain your changes.
This PR allows for the attributes for KMeans to be read-only. When
update_centers=False,lloyd_iter_chunked_densedoes not actually updatecenters_new. This is the case whenKMeans.predictends up calling_labels_inertia:scikit-learn/sklearn/cluster/_kmeans.py
Line 808 in feaf382
Any other comments?
Side note: I think the new test in this PR for read only attributes can become a common test.