FEA metadata routing for `StackingClassifier` and `StackingRegressor` #28701

StefanieSenger · 2024-03-26T14:36:49Z

Reference Issues/PRs

towards #22893
closes #18028

What does this implement/fix? Explain your changes.

Adds metadata routing to StackingClassifier and StackingRegressor.

Any other comments?

I wasn't sure if we want to route metadata within the predict method. It's already implemented in _BaseStacking.predict(self, X, **predict_params), but without needing to set a request for it. What is the preferable way here?
Also there is an issue whenever RidgeCV (the default) is the final_estimator: We get a RecursionError, because it gets hung in _metadata_requests.py. I will continue to invest about this. Update: This is a bug in the routing mechanism of RidgeCV and it's unrelated to this PR. I've opened a PR to fix it: FIX RecursionError bug with metadata routing in metaestimators with scoring #28712

@adrinjalali @OmarManzoor @glemaitre, do you want to have a look?

github-actions · 2024-03-26T14:38:05Z

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 01b8efe. Link to the linter CI: here}

adrinjalali · 2024-04-09T15:10:22Z

sklearn/ensemble/_stacking.py

+            .. deprecated:: 1.5
+                `sample_weight` is deprecated in 1.5 and will be removed in 1.7.


in this class it's not deprecated though, it's simply removed.

Oh yes, true.

sklearn/ensemble/_stacking.py

adrinjalali · 2024-04-09T15:57:48Z

sklearn/ensemble/_stacking.py

-    def fit_transform(self, X, y, sample_weight=None):
+    def fit_transform(self, X, y, sample_weight=None, **fit_params):


I think we should deprecate positional sample_weight here as well.

Yes, we should. Otherwise a sample_weight passed as a fit_param into the routing would mistakenly be passed through the old way outside of the routing.

adrinjalali · 2024-04-09T16:04:25Z

sklearn/tests/metadata_routing_common.py

+        record_metadata_not_default(
+            self, "predict_proba", sample_weight=sample_weight, metadata=metadata
+        )
+        return np.asarray([[0.0, 1.0]] * len(X))


where is this used?

In StackingClassifier, there's a stack_method param, whiches value can be set to predict_proba as well. In StackingRregessor it's just hardcoded as predict.

Co-authored-by: Adrin Jalali <adrin.jalali@gmail.com>

StefanieSenger

@adrinjalali
Thanks for reviewing. I went through it and improved regarding your comments.

StefanieSenger · 2024-04-10T08:39:08Z

sklearn/ensemble/_stacking.py

+            .. deprecated:: 1.5
+                `sample_weight` is deprecated in 1.5 and will be removed in 1.7.


Oh yes, true.

StefanieSenger · 2024-04-10T12:39:18Z

sklearn/ensemble/_stacking.py

-    def fit_transform(self, X, y, sample_weight=None):
+    def fit_transform(self, X, y, sample_weight=None, **fit_params):


Yes, we should. Otherwise a sample_weight passed as a fit_param into the routing would mistakenly be passed through the old way outside of the routing.

StefanieSenger · 2024-04-11T09:05:01Z

sklearn/tests/metadata_routing_common.py

+        record_metadata_not_default(
+            self, "predict_proba", sample_weight=sample_weight, metadata=metadata
+        )
+        return np.asarray([[0.0, 1.0]] * len(X))


In StackingClassifier, there's a stack_method param, whiches value can be set to predict_proba as well. In StackingRregessor it's just hardcoded as predict.

StefanieSenger · 2024-04-18T14:25:45Z

So, the RecursionError bug that is about to be fixed in #28712 is also appearing here.

adrinjalali · 2024-04-29T15:20:26Z

@StefanieSenger since #28712 is merged, wanna merge with main?

StefanieSenger · 2024-04-30T12:29:29Z

I have merged main into it and all tests now pass, @adrinjalali

OmarManzoor

Thanks for the PR @StefanieSenger. Just a few minor comments otherwise this looks good.

sklearn/ensemble/_stacking.py

sklearn/ensemble/tests/test_stacking.py

Co-authored-by: Omar Salman <omar.salman@arbisoft.com>

StefanieSenger

Thanks for reviewing @OmarManzoor. I have committed those changes. :)

OmarManzoor

A further suggestion

sklearn/ensemble/tests/test_stacking.py

Co-authored-by: Omar Salman <omar.salman@arbisoft.com>

StefanieSenger · 2024-05-08T13:32:24Z

A further suggestion

Oh, that's wonderful, thank you, @OmarManzoor :)

OmarManzoor

LGTM. Thanks @StefanieSenger. The versions will just need to be changed to 1.6 now that the branch for 1.5 has already been separated.

OmarManzoor · 2024-05-13T06:08:44Z

doc/whats_new/v1.5.rst

@@ -139,6 +139,11 @@ more details.
  transformers' ``fit`` and ``fit_transform``. :pr:`28205` by :user:`Stefanie
  Senger <StefanieSenger>`.

+- |Feature| :class:`ensemble.StackingClassifier` and


I think this might need to move to 1.6 now.

I've moved it :)

sklearn/ensemble/_stacking.py

Co-authored-by: Omar Salman <omar.salman@arbisoft.com>

metadata routing for stackingclassifier and stackingregressor

5f8d871

github-actions bot added the module:ensemble label Mar 26, 2024

StefanieSenger added 3 commits March 26, 2024 16:07

little fixes

db318b7

fix docstring

5706e19

correct deprecation version

b97f5af

glemaitre self-requested a review April 8, 2024 13:07

adrinjalali reviewed Apr 9, 2024

View reviewed changes

Apply suggestions from code review

109bbd5

Co-authored-by: Adrin Jalali <adrin.jalali@gmail.com>

StefanieSenger commented Apr 11, 2024

View reviewed changes

StefanieSenger and others added 7 commits April 11, 2024 11:07

changes after review

98bca6f

Merge branch 'main' into routing_Stacking

f96b7d3

fix CI

aaf8df1

buggy routing for predict

2602fb1

only test last entry in registry for final estimator

733bcac

fix docstring

c6762eb

fix docstring

4c271d2

Merge branch 'main' into routing_Stacking

c260459

OmarManzoor reviewed May 6, 2024

View reviewed changes

sklearn/ensemble/_stacking.py Outdated Show resolved Hide resolved

sklearn/ensemble/_stacking.py Show resolved Hide resolved

sklearn/ensemble/tests/test_stacking.py Outdated Show resolved Hide resolved

Apply suggestions from code review

b8dd378

Co-authored-by: Omar Salman <omar.salman@arbisoft.com>

StefanieSenger commented May 8, 2024

View reviewed changes

Merge branch 'main' into routing_Stacking

cc9a682

OmarManzoor reviewed May 8, 2024

View reviewed changes

sklearn/ensemble/tests/test_stacking.py Outdated Show resolved Hide resolved

StefanieSenger and others added 2 commits May 8, 2024 14:43

Update sklearn/ensemble/tests/test_stacking.py

38a3957

Co-authored-by: Omar Salman <omar.salman@arbisoft.com>

simplify test

4a8ea49

OmarManzoor approved these changes May 13, 2024

View reviewed changes

StefanieSenger and others added 4 commits May 13, 2024 08:46

Apply suggestions from code review

75c23d3

Co-authored-by: Omar Salman <omar.salman@arbisoft.com>

update changelog

ba7359e

delete line

65322ac

Merge branch 'main' into routing_Stacking

01b8efe

adrinjalali approved these changes May 13, 2024

View reviewed changes

adrinjalali merged commit 61281cf into scikit-learn:main May 13, 2024
30 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

FEA metadata routing for `StackingClassifier` and `StackingRegressor` #28701

FEA metadata routing for `StackingClassifier` and `StackingRegressor` #28701

StefanieSenger commented Mar 26, 2024 •

edited

github-actions bot commented Mar 26, 2024 •

edited

adrinjalali Apr 9, 2024

StefanieSenger Apr 10, 2024

adrinjalali Apr 9, 2024

StefanieSenger Apr 10, 2024

adrinjalali Apr 9, 2024

StefanieSenger Apr 11, 2024

StefanieSenger left a comment

StefanieSenger Apr 10, 2024

StefanieSenger Apr 10, 2024

StefanieSenger Apr 11, 2024

StefanieSenger commented Apr 18, 2024

adrinjalali commented Apr 29, 2024

StefanieSenger commented Apr 30, 2024

OmarManzoor left a comment

StefanieSenger left a comment

OmarManzoor left a comment

StefanieSenger commented May 8, 2024

OmarManzoor left a comment

OmarManzoor May 13, 2024

StefanieSenger May 13, 2024

		.. deprecated:: 1.5
		`sample_weight` is deprecated in 1.5 and will be removed in 1.7.

		def fit_transform(self, X, y, sample_weight=None):
		def fit_transform(self, X, y, sample_weight=None, **fit_params):

FEA metadata routing for StackingClassifier and StackingRegressor #28701

FEA metadata routing for StackingClassifier and StackingRegressor #28701

Conversation

StefanieSenger commented Mar 26, 2024 • edited

Reference Issues/PRs

What does this implement/fix? Explain your changes.

Any other comments?

github-actions bot commented Mar 26, 2024 • edited

✔️ Linting Passed

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

StefanieSenger left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

StefanieSenger commented Apr 18, 2024

adrinjalali commented Apr 29, 2024

StefanieSenger commented Apr 30, 2024

OmarManzoor left a comment

Choose a reason for hiding this comment

StefanieSenger left a comment

Choose a reason for hiding this comment

OmarManzoor left a comment

Choose a reason for hiding this comment

StefanieSenger commented May 8, 2024

OmarManzoor left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

FEA metadata routing for `StackingClassifier` and `StackingRegressor` #28701

FEA metadata routing for `StackingClassifier` and `StackingRegressor` #28701

StefanieSenger commented Mar 26, 2024 •

edited

github-actions bot commented Mar 26, 2024 •

edited