[MRG] Standardize sample weights validation in DummyClassifier #15510
Conversation
|
Looks good pending CI evaluation |
| @@ -141,7 +142,10 @@ def fit(self, X, y, sample_weight=None): | |||
|
|
|||
| self.n_outputs_ = y.shape[1] | |||
|
|
|||
| check_consistent_length(X, y, sample_weight) | |||
| check_consistent_length(X, y) | |||
There was a problem hiding this comment.
I think we might still want it here, not strong opinion, though...
There was a problem hiding this comment.
even though it's checked as part of the added validation?
There was a problem hiding this comment.
+1 to avoid redundant checks (even if it doesn't cost much). _check_sample_weight should yield better error messages
| @@ -141,7 +142,10 @@ def fit(self, X, y, sample_weight=None): | |||
|
|
|||
| self.n_outputs_ = y.shape[1] | |||
|
|
|||
| check_consistent_length(X, y, sample_weight) | |||
| check_consistent_length(X, y) | |||
There was a problem hiding this comment.
+1 to avoid redundant checks (even if it doesn't cost much). _check_sample_weight should yield better error messages
sklearn/tests/test_dummy.py
Outdated
| clf = DummyClassifier().fit(X, y, sample_weight) | ||
| assert_array_almost_equal(clf.class_prior_, [0.2 / 1.2, 1. / 1.2]) | ||
|
|
||
| sample_weight = np.random.rand(3, 1) |
There was a problem hiding this comment.
For anything below this line, I'm not sure there is much sense in adding sample_weight tests for each estimator (unless they have some special behavior there) as they would be quite redundant, instead they should be enforced by common tests in sklearn/utils/estimator_checks.py (and for a number of below point they may already be).
So I would remove the below test. The check specific to DummyClassifier above is quite nice though.
There was a problem hiding this comment.
Thanks for the review!
We removed the test below.
|
Something went wrong with the merge of master here. The diff should only show the changed code... |
(cherry picked from commit 95929b9)
|
I'm Fanny's pair from the workshop. I'm not sure what happened with the merge from master. I have just branched off of master again and cherry-picked the 2 commits for ease of fixing everything. I can either push to a new branch and create a new PR or force push my current branch to overwrite the changes. Do the maintainers have a preference? |
Co-authored-by: Sallie Walecka <[email protected]> (cherry picked from commit e6bced8)
|
in this case I think force-pushing is fine if all the comments are addressed. |
dfa0c46 to
caa7871
Compare
|
Looks like CI is failing due to some warning coming from the test. However, the test is identical to the test case above it (apart from stratified sampling), so I just ended up deleting it. Pushing changes now. |
|
@amueller everything should be good now |
| check_consistent_length(X, y) | ||
|
|
||
| if sample_weight is not None: | ||
| sample_weight = _check_sample_weight(sample_weight, X) |
There was a problem hiding this comment.
Actually it looks like several PRs were done for this estimator as this was added on the line below in #15505. Please remove the above 3 lines.
Otherwise LGTM.
There was a problem hiding this comment.
Is it worth merging at this point? Looks like the only changes left would be one linting issue and removing sample weight from check_consistent_length. Does it make more sense to close the PR instead?
|
@cmarmo wondering if there still needs to be work done on this one? |
…t-learn#15510) Co-authored-by: Sallie Walecka <[email protected]>
Reference Issues/PRs
Partially addresses #15358 for DummyClassifier
What does this implement/fix? Explain your changes.
Replaces custom validation logic with standardized method
utils.validation._check_sample_weight(relatively newly introduced).Any other comments?