From c213ea5063776f67a60568714d2a8177721b4004 Mon Sep 17 00:00:00 2001
From: Miguel Silva <miguelcesilva@gmail.com>
Date: Sun, 24 Mar 2024 21:31:07 +0000
Subject: [PATCH 1/8] Add n_features to RFECV.cv_results_

---
 doc/whats_new/v1.5.rst                        | 10 ++++-
 .../plot_rfe_with_cross_validation.py         |  9 ++--
 sklearn/feature_selection/_rfe.py             | 24 ++++++++---
 sklearn/feature_selection/tests/test_rfe.py   | 43 ++++++++++++++++++-
 4 files changed, 72 insertions(+), 14 deletions(-)

diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
index bd03cc743f76e..4d41209e1ee8e 100644
--- a/doc/whats_new/v1.5.rst
+++ b/doc/whats_new/v1.5.rst
@@ -188,6 +188,14 @@ Changelog
   :pr:`28085` by :user:`Neto Menoci <netomenoci>` and
   :user:`Florin Andrei <FlorinAndrei>`.
 
+- |Enhancement| :class:`feature_selection.RFE`  now have the `n_features_fitted_`
+  attribute after `_fit`.
+  :pr:`43745` by :user:`Miguel Silva <miguelcsilva>`.
+
+- |Feature| :class:`feature_selection.RFECV` adds a new key to its `cv_results_`
+  `dict`, containing an `np.array` with the number of features fitted at each step.
+  :pr:`43745` by :user:`Miguel Silva <miguelcsilva>`.
+
 :mod:`sklearn.impute`
 .....................
 
@@ -298,7 +306,7 @@ Changelog
   :func:`preprocessing.quantile_transform` now supports disabling
   subsampling explicitly.
   :pr:`27636` by :user:`Ralph Urlus <rurlus>`.
-  
+
 :mod:`sklearn.tree`
 ...................
 
diff --git a/examples/feature_selection/plot_rfe_with_cross_validation.py b/examples/feature_selection/plot_rfe_with_cross_validation.py
index 693e21fe21787..6e4a8ae0ee8c5 100644
--- a/examples/feature_selection/plot_rfe_with_cross_validation.py
+++ b/examples/feature_selection/plot_rfe_with_cross_validation.py
@@ -66,15 +66,16 @@
 # ---------------------------------------------------
 
 import matplotlib.pyplot as plt
+import pandas as pd
 
-n_scores = len(rfecv.cv_results_["mean_test_score"])
+cv_results = pd.DataFrame(rfecv.cv_results_)
 plt.figure()
 plt.xlabel("Number of features selected")
 plt.ylabel("Mean test accuracy")
 plt.errorbar(
-    range(min_features_to_select, n_scores + min_features_to_select),
-    rfecv.cv_results_["mean_test_score"],
-    yerr=rfecv.cv_results_["std_test_score"],
+    x=cv_results["n_features"],
+    y=cv_results["mean_test_score"],
+    yerr=cv_results["std_test_score"],
 )
 plt.title("Recursive Feature Elimination \nwith correlated features")
 plt.show()
diff --git a/sklearn/feature_selection/_rfe.py b/sklearn/feature_selection/_rfe.py
index e85dc8f623596..2094a314ed326 100644
--- a/sklearn/feature_selection/_rfe.py
+++ b/sklearn/feature_selection/_rfe.py
@@ -153,6 +153,9 @@ class RFE(_RoutingNotSupportedMixin, SelectorMixin, MetaEstimatorMixin, BaseEsti
     support_ : ndarray of shape (n_features,)
         The mask of selected features.
 
+    n_features_fitted_: ndarray of shape (n_features_in_ // step [+1],)
+        Number of features used for fitting at each step.
+
     See Also
     --------
     RFECV : Recursive feature elimination with built-in cross-validated
@@ -294,6 +297,7 @@ def _fit(self, X, y, step_score=None, **fit_params):
 
         support_ = np.ones(n_features, dtype=bool)
         ranking_ = np.ones(n_features, dtype=int)
+        self.n_features_fitted_ = []
 
         if step_score:
             self.scores_ = []
@@ -329,6 +333,7 @@ def _fit(self, X, y, step_score=None, **fit_params):
             # that have not been eliminated yet
             if step_score:
                 self.scores_.append(step_score(estimator, features))
+                self.n_features_fitted_.append(len(features))
             support_[features[ranks][:threshold]] = False
             ranking_[np.logical_not(support_)] += 1
 
@@ -340,9 +345,11 @@ def _fit(self, X, y, step_score=None, **fit_params):
         # Compute step score when only n_features_to_select features left
         if step_score:
             self.scores_.append(step_score(self.estimator_, features))
+            self.n_features_fitted_.append(len(features))
         self.n_features_ = support_.sum()
         self.support_ = support_
         self.ranking_ = ranking_
+        self.n_features_fitted_ = np.array(self.n_features_fitted_)
 
         return self
 
@@ -581,6 +588,9 @@ class RFECV(RFE):
         std_test_score : ndarray of shape (n_subsets_of_features,)
             Standard deviation of scores over the folds.
 
+        n_features : ndarray of shape (n_subsets_of_features,)
+            Number of features used at each step.
+
         .. versionadded:: 1.0
 
     n_features_ : int
@@ -758,6 +768,7 @@ def fit(self, X, y, groups=None):
             for train, test in cv.split(X, y, groups)
         )
 
+        n_features_fitted = rfe.n_features_fitted_
         scores = np.array(scores)
         scores_sum = np.sum(scores, axis=0)
         scores_sum_rev = scores_sum[::-1]
@@ -786,11 +797,10 @@ def fit(self, X, y, groups=None):
 
         # reverse to stay consistent with before
         scores_rev = scores[:, ::-1]
-        self.cv_results_ = {}
-        self.cv_results_["mean_test_score"] = np.mean(scores_rev, axis=0)
-        self.cv_results_["std_test_score"] = np.std(scores_rev, axis=0)
-
-        for i in range(scores.shape[0]):
-            self.cv_results_[f"split{i}_test_score"] = scores_rev[i]
-
+        self.cv_results_ = {
+            "mean_test_score": np.mean(scores_rev, axis=0),
+            "std_test_score": np.std(scores_rev, axis=0),
+            **{f"split{i}_test_score": scores_rev[i] for i in range(scores.shape[0])},
+            "n_features": np.flip(n_features_fitted),
+        }
         return self
diff --git a/sklearn/feature_selection/tests/test_rfe.py b/sklearn/feature_selection/tests/test_rfe.py
index e3edb0e7b5d21..6c9e0e87c96c7 100644
--- a/sklearn/feature_selection/tests/test_rfe.py
+++ b/sklearn/feature_selection/tests/test_rfe.py
@@ -11,7 +11,7 @@
 from sklearn.base import BaseEstimator, ClassifierMixin
 from sklearn.compose import TransformedTargetRegressor
 from sklearn.cross_decomposition import CCA, PLSCanonical, PLSRegression
-from sklearn.datasets import load_iris, make_friedman1
+from sklearn.datasets import load_iris, make_classification, make_friedman1
 from sklearn.ensemble import RandomForestClassifier
 from sklearn.feature_selection import RFE, RFECV
 from sklearn.impute import SimpleImputer
@@ -537,7 +537,8 @@ def test_rfecv_std_and_mean(global_random_seed):
 
     rfecv = RFECV(estimator=SVC(kernel="linear"))
     rfecv.fit(X, y)
-    n_split_keys = len(rfecv.cv_results_) - 2
+    n_non_split_keys = 3
+    n_split_keys = len(rfecv.cv_results_) - n_non_split_keys
     split_keys = [f"split{i}_test_score" for i in range(n_split_keys)]
 
     cv_scores = np.asarray([rfecv.cv_results_[key] for key in split_keys])
@@ -548,6 +549,44 @@ def test_rfecv_std_and_mean(global_random_seed):
     assert_allclose(rfecv.cv_results_["std_test_score"], expected_std)
 
 
+@pytest.mark.parametrize(
+    ["min_features_to_select", "n_features", "step", "cv_results_n_features"],
+    [
+        [1, 4, 1, np.array([1, 2, 3, 4])],
+        [1, 5, 1, np.array([1, 2, 3, 4, 5])],
+        [1, 4, 2, np.array([1, 2, 4])],
+        [1, 5, 2, np.array([1, 3, 5])],
+        [1, 4, 3, np.array([1, 4])],
+        [1, 5, 3, np.array([1, 2, 5])],
+        [1, 4, 4, np.array([1, 4])],
+        [1, 5, 4, np.array([1, 5])],
+        [4, 4, 2, np.array([4])],
+        [4, 5, 1, np.array([4, 5])],
+        [4, 5, 2, np.array([4, 5])],
+    ],
+)
+def test_rfecv_cv_results_n_features(
+    min_features_to_select,
+    n_features,
+    step,
+    cv_results_n_features,
+):
+    X, y = make_classification(
+        n_samples=20, n_features=n_features, n_informative=n_features, n_redundant=0
+    )
+    rfecv = RFECV(
+        estimator=SVC(kernel="linear"),
+        step=step,
+        min_features_to_select=min_features_to_select,
+    )
+    rfecv.fit(X, y)
+    assert_array_equal(rfecv.cv_results_["n_features"], cv_results_n_features)
+    assert all(
+        len(value) == len(rfecv.cv_results_["n_features"])
+        for value in rfecv.cv_results_.values()
+    )
+
+
 @pytest.mark.parametrize("ClsRFE", [RFE, RFECV])
 def test_multioutput(ClsRFE):
     X = np.random.normal(size=(10, 3))

From 8a57ab3243297f2f31fafd97d68df1a7f467cd62 Mon Sep 17 00:00:00 2001
From: Miguel Silva <miguelcesilva@gmail.com>
Date: Mon, 25 Mar 2024 18:03:34 +0000
Subject: [PATCH 2/8] Fix pr in changelog

---
 doc/whats_new/v1.5.rst | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
index 4d41209e1ee8e..8d8e3128d969f 100644
--- a/doc/whats_new/v1.5.rst
+++ b/doc/whats_new/v1.5.rst
@@ -190,11 +190,11 @@ Changelog
 
 - |Enhancement| :class:`feature_selection.RFE`  now have the `n_features_fitted_`
   attribute after `_fit`.
-  :pr:`43745` by :user:`Miguel Silva <miguelcsilva>`.
+  :pr:`28670` by :user:`Miguel Silva <miguelcsilva>`.
 
 - |Feature| :class:`feature_selection.RFECV` adds a new key to its `cv_results_`
   `dict`, containing an `np.array` with the number of features fitted at each step.
-  :pr:`43745` by :user:`Miguel Silva <miguelcsilva>`.
+  :pr:`28670` by :user:`Miguel Silva <miguelcsilva>`.
 
 :mod:`sklearn.impute`
 .....................

From 2fa460d20902addd72628f732f1867cde3dee10a Mon Sep 17 00:00:00 2001
From: jeremie du boisberranger <jeremiedbb@yahoo.fr>
Date: Tue, 26 Mar 2024 00:02:37 +0100
Subject: [PATCH 3/8] make new attr non public + simplify argmax

---
 doc/whats_new/v1.5.rst                      |  9 ++---
 sklearn/feature_selection/_rfe.py           | 40 +++++++++------------
 sklearn/feature_selection/tests/test_rfe.py |  5 +--
 3 files changed, 20 insertions(+), 34 deletions(-)

diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
index 4d41209e1ee8e..e79006c3a0840 100644
--- a/doc/whats_new/v1.5.rst
+++ b/doc/whats_new/v1.5.rst
@@ -188,12 +188,9 @@ Changelog
   :pr:`28085` by :user:`Neto Menoci <netomenoci>` and
   :user:`Florin Andrei <FlorinAndrei>`.
 
-- |Enhancement| :class:`feature_selection.RFE`  now have the `n_features_fitted_`
-  attribute after `_fit`.
-  :pr:`43745` by :user:`Miguel Silva <miguelcsilva>`.
-
-- |Feature| :class:`feature_selection.RFECV` adds a new key to its `cv_results_`
-  `dict`, containing an `np.array` with the number of features fitted at each step.
+- |Enhancement| The `cv_results_` attribute of :class:`feature_selection.RFECV`has
+  a new key, `n_features`, containing an array with the number of features selected
+  at each step.
   :pr:`43745` by :user:`Miguel Silva <miguelcsilva>`.
 
 :mod:`sklearn.impute`
diff --git a/sklearn/feature_selection/_rfe.py b/sklearn/feature_selection/_rfe.py
index 2094a314ed326..bf8dff62dec7a 100644
--- a/sklearn/feature_selection/_rfe.py
+++ b/sklearn/feature_selection/_rfe.py
@@ -32,7 +32,8 @@ def _rfe_single_fit(rfe, estimator, X, y, train, test, scorer):
     """
     X_train, y_train = _safe_split(estimator, X, y, train)
     X_test, y_test = _safe_split(estimator, X, y, test, train)
-    return rfe._fit(
+
+    rfe._fit(
         X_train,
         y_train,
         lambda estimator, features: _score(
@@ -43,7 +44,9 @@ def _rfe_single_fit(rfe, estimator, X, y, train, test, scorer):
             scorer,
             score_params=None,
         ),
-    ).scores_
+    )
+
+    return rfe.scores_, rfe.n_features_selected_
 
 
 def _estimator_has(attr):
@@ -153,9 +156,6 @@ class RFE(_RoutingNotSupportedMixin, SelectorMixin, MetaEstimatorMixin, BaseEsti
     support_ : ndarray of shape (n_features,)
         The mask of selected features.
 
-    n_features_fitted_: ndarray of shape (n_features_in_ // step [+1],)
-        Number of features used for fitting at each step.
-
     See Also
     --------
     RFECV : Recursive feature elimination with built-in cross-validated
@@ -297,9 +297,9 @@ def _fit(self, X, y, step_score=None, **fit_params):
 
         support_ = np.ones(n_features, dtype=bool)
         ranking_ = np.ones(n_features, dtype=int)
-        self.n_features_fitted_ = []
 
         if step_score:
+            self.n_features_selected_ = []
             self.scores_ = []
 
         # Elimination
@@ -332,8 +332,8 @@ def _fit(self, X, y, step_score=None, **fit_params):
             # because 'estimator' must use features
             # that have not been eliminated yet
             if step_score:
+                self.n_features_selected_.append(len(features))
                 self.scores_.append(step_score(estimator, features))
-                self.n_features_fitted_.append(len(features))
             support_[features[ranks][:threshold]] = False
             ranking_[np.logical_not(support_)] += 1
 
@@ -344,12 +344,11 @@ def _fit(self, X, y, step_score=None, **fit_params):
 
         # Compute step score when only n_features_to_select features left
         if step_score:
+            self.n_features_selected_.append(len(features))
             self.scores_.append(step_score(self.estimator_, features))
-            self.n_features_fitted_.append(len(features))
         self.n_features_ = support_.sum()
         self.support_ = support_
         self.ranking_ = ranking_
-        self.n_features_fitted_ = np.array(self.n_features_fitted_)
 
         return self
 
@@ -728,12 +727,6 @@ def fit(self, X, y, groups=None):
         # Initialization
         cv = check_cv(self.cv, y, classifier=is_classifier(self.estimator))
         scorer = check_scoring(self.estimator, scoring=self.scoring)
-        n_features = X.shape[1]
-
-        if 0.0 < self.step < 1.0:
-            step = int(max(1, self.step * n_features))
-        else:
-            step = int(self.step)
 
         # Build an RFE object, which will evaluate and score each possible
         # feature count, down to self.min_features_to_select
@@ -763,19 +756,18 @@ def fit(self, X, y, groups=None):
             parallel = Parallel(n_jobs=self.n_jobs)
             func = delayed(_rfe_single_fit)
 
-        scores = parallel(
+        scores_features = parallel(
             func(rfe, self.estimator, X, y, train, test, scorer)
             for train, test in cv.split(X, y, groups)
         )
+        scores, n_features_per_iter = zip(*scores_features)
 
-        n_features_fitted = rfe.n_features_fitted_
+        n_features_per_iter_rev = np.array(n_features_per_iter[0])[::-1]
         scores = np.array(scores)
-        scores_sum = np.sum(scores, axis=0)
-        scores_sum_rev = scores_sum[::-1]
-        argmax_idx = len(scores_sum) - np.argmax(scores_sum_rev) - 1
-        n_features_to_select = max(
-            n_features - (argmax_idx * step), self.min_features_to_select
-        )
+
+        # Reverse order such that lowest number of features is selected in case of tie.
+        scores_sum_rev = np.sum(scores, axis=0)[::-1]
+        n_features_to_select = n_features_per_iter_rev[np.argmax(scores_sum_rev)]
 
         # Re-execute an elimination with best_k over the whole set
         rfe = RFE(
@@ -801,6 +793,6 @@ def fit(self, X, y, groups=None):
             "mean_test_score": np.mean(scores_rev, axis=0),
             "std_test_score": np.std(scores_rev, axis=0),
             **{f"split{i}_test_score": scores_rev[i] for i in range(scores.shape[0])},
-            "n_features": np.flip(n_features_fitted),
+            "n_features": n_features_per_iter_rev,
         }
         return self
diff --git a/sklearn/feature_selection/tests/test_rfe.py b/sklearn/feature_selection/tests/test_rfe.py
index 6c9e0e87c96c7..01c6194493ab6 100644
--- a/sklearn/feature_selection/tests/test_rfe.py
+++ b/sklearn/feature_selection/tests/test_rfe.py
@@ -537,10 +537,7 @@ def test_rfecv_std_and_mean(global_random_seed):
 
     rfecv = RFECV(estimator=SVC(kernel="linear"))
     rfecv.fit(X, y)
-    n_non_split_keys = 3
-    n_split_keys = len(rfecv.cv_results_) - n_non_split_keys
-    split_keys = [f"split{i}_test_score" for i in range(n_split_keys)]
-
+    split_keys = [key for key in rfecv.cv_results_.keys() if "split" in key]
     cv_scores = np.asarray([rfecv.cv_results_[key] for key in split_keys])
     expected_mean = np.mean(cv_scores, axis=0)
     expected_std = np.std(cv_scores, axis=0)

From 29ff8100c99a59de1a5c73ee00ed43447b7cc09c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?J=C3=A9r=C3=A9mie=20du=20Boisberranger?=
 <jeremie@probabl.ai>
Date: Tue, 26 Mar 2024 11:13:11 +0100
Subject: [PATCH 4/8] cln

---
 doc/whats_new/v1.5.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
index e487809e7a2b2..32a6fc280d531 100644
--- a/doc/whats_new/v1.5.rst
+++ b/doc/whats_new/v1.5.rst
@@ -188,7 +188,7 @@ Changelog
   :pr:`28085` by :user:`Neto Menoci <netomenoci>` and
   :user:`Florin Andrei <FlorinAndrei>`.
 
-- |Enhancement| The `cv_results_` attribute of :class:`feature_selection.RFECV`has
+- |Enhancement| The `cv_results_` attribute of :class:`feature_selection.RFECV` has
   a new key, `n_features`, containing an array with the number of features selected
   at each step.
   :pr:`28670` by :user:`Miguel Silva <miguelcsilva>`.

From 4fca941eec0748c5c79ff3ea71ae30663211d16f Mon Sep 17 00:00:00 2001
From: jeremie du boisberranger <jeremiedbb@yahoo.fr>
Date: Tue, 26 Mar 2024 11:16:04 +0100
Subject: [PATCH 5/8] add meson-python to requirements

---
 build_tools/azure/python_nogil_lock.txt         | 2 +-
 build_tools/azure/python_nogil_requirements.txt | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/build_tools/azure/python_nogil_lock.txt b/build_tools/azure/python_nogil_lock.txt
index 85988263b94b4..9dba84dfb619e 100644
--- a/build_tools/azure/python_nogil_lock.txt
+++ b/build_tools/azure/python_nogil_lock.txt
@@ -55,7 +55,7 @@ scipy==1.9.3
     # via -r /scikit-learn/build_tools/azure/python_nogil_requirements.txt
 six==1.16.0
     # via python-dateutil
-threadpoolctl==3.3.0
+threadpoolctl==3.4.0
     # via -r /scikit-learn/build_tools/azure/python_nogil_requirements.txt
 tomli==2.0.1
     # via pytest
diff --git a/build_tools/azure/python_nogil_requirements.txt b/build_tools/azure/python_nogil_requirements.txt
index 970059ede81aa..959b5ace581a8 100644
--- a/build_tools/azure/python_nogil_requirements.txt
+++ b/build_tools/azure/python_nogil_requirements.txt
@@ -13,3 +13,4 @@ joblib
 threadpoolctl
 pytest
 pytest-xdist
+meson-python

From b777435116652081c4180bb78457bd665ebc1086 Mon Sep 17 00:00:00 2001
From: Miguel Silva <miguelcesilva@gmail.com>
Date: Tue, 26 Mar 2024 21:06:57 +0000
Subject: [PATCH 6/8] Rename private vars

---
 sklearn/feature_selection/_rfe.py | 21 ++++++++++-----------
 1 file changed, 10 insertions(+), 11 deletions(-)

diff --git a/sklearn/feature_selection/_rfe.py b/sklearn/feature_selection/_rfe.py
index bf8dff62dec7a..c3ddf168cf5c5 100644
--- a/sklearn/feature_selection/_rfe.py
+++ b/sklearn/feature_selection/_rfe.py
@@ -46,7 +46,7 @@ def _rfe_single_fit(rfe, estimator, X, y, train, test, scorer):
         ),
     )
 
-    return rfe.scores_, rfe.n_features_selected_
+    return rfe.step_scores_, rfe.step_n_features_
 
 
 def _estimator_has(attr):
@@ -267,10 +267,9 @@ def fit(self, X, y, **fit_params):
         return self._fit(X, y, **fit_params)
 
     def _fit(self, X, y, step_score=None, **fit_params):
-        # Parameter step_score controls the calculation of self.scores_
-        # step_score is not exposed to users
-        # and is used when implementing RFECV
-        # self.scores_ will not be calculated when calling _fit through fit
+        # Parameter step_score controls the calculation of self.step_scores_
+        # step_score is not exposed to users and is used when implementing RFECV
+        # self.step_scores_ will not be calculated when calling _fit through fit
 
         X, y = self._validate_data(
             X,
@@ -299,8 +298,8 @@ def _fit(self, X, y, step_score=None, **fit_params):
         ranking_ = np.ones(n_features, dtype=int)
 
         if step_score:
-            self.n_features_selected_ = []
-            self.scores_ = []
+            self.step_n_features_ = []
+            self.step_scores_ = []
 
         # Elimination
         while np.sum(support_) > n_features_to_select:
@@ -332,8 +331,8 @@ def _fit(self, X, y, step_score=None, **fit_params):
             # because 'estimator' must use features
             # that have not been eliminated yet
             if step_score:
-                self.n_features_selected_.append(len(features))
-                self.scores_.append(step_score(estimator, features))
+                self.step_n_features_.append(len(features))
+                self.step_scores_.append(step_score(estimator, features))
             support_[features[ranks][:threshold]] = False
             ranking_[np.logical_not(support_)] += 1
 
@@ -344,8 +343,8 @@ def _fit(self, X, y, step_score=None, **fit_params):
 
         # Compute step score when only n_features_to_select features left
         if step_score:
-            self.n_features_selected_.append(len(features))
-            self.scores_.append(step_score(self.estimator_, features))
+            self.step_n_features_.append(len(features))
+            self.step_scores_.append(step_score(self.estimator_, features))
         self.n_features_ = support_.sum()
         self.support_ = support_
         self.ranking_ = ranking_

From 3dcbb8ae2960d0b588f194b73a82652faf09496b Mon Sep 17 00:00:00 2001
From: jeremie du boisberranger <jeremiedbb@yahoo.fr>
Date: Wed, 27 Mar 2024 14:57:53 +0100
Subject: [PATCH 7/8] a little bit more renaming

---
 sklearn/feature_selection/_rfe.py | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/sklearn/feature_selection/_rfe.py b/sklearn/feature_selection/_rfe.py
index bf8dff62dec7a..34bcd54241d07 100644
--- a/sklearn/feature_selection/_rfe.py
+++ b/sklearn/feature_selection/_rfe.py
@@ -28,7 +28,7 @@
 
 def _rfe_single_fit(rfe, estimator, X, y, train, test, scorer):
     """
-    Return the score for a fit across one fold.
+    Return the score and n_features per step for a fit across one fold.
     """
     X_train, y_train = _safe_split(estimator, X, y, train)
     X_test, y_test = _safe_split(estimator, X, y, test, train)
@@ -760,14 +760,14 @@ def fit(self, X, y, groups=None):
             func(rfe, self.estimator, X, y, train, test, scorer)
             for train, test in cv.split(X, y, groups)
         )
-        scores, n_features_per_iter = zip(*scores_features)
+        scores, step_n_features = zip(*scores_features)
 
-        n_features_per_iter_rev = np.array(n_features_per_iter[0])[::-1]
+        step_n_features_rev = np.array(step_n_features[0])[::-1]
         scores = np.array(scores)
 
         # Reverse order such that lowest number of features is selected in case of tie.
         scores_sum_rev = np.sum(scores, axis=0)[::-1]
-        n_features_to_select = n_features_per_iter_rev[np.argmax(scores_sum_rev)]
+        n_features_to_select = step_n_features_rev[np.argmax(scores_sum_rev)]
 
         # Re-execute an elimination with best_k over the whole set
         rfe = RFE(
@@ -793,6 +793,6 @@ def fit(self, X, y, groups=None):
             "mean_test_score": np.mean(scores_rev, axis=0),
             "std_test_score": np.std(scores_rev, axis=0),
             **{f"split{i}_test_score": scores_rev[i] for i in range(scores.shape[0])},
-            "n_features": n_features_per_iter_rev,
+            "n_features": step_n_features_rev,
         }
         return self

From 24be84eb104d4dd665ee70e286ca5ea44e9d9a85 Mon Sep 17 00:00:00 2001
From: jeremie du boisberranger <jeremiedbb@yahoo.fr>
Date: Wed, 27 Mar 2024 15:00:43 +0100
Subject: [PATCH 8/8] oops

---
 build_tools/azure/python_nogil_lock.txt         | 2 +-
 build_tools/azure/python_nogil_requirements.txt | 1 -
 2 files changed, 1 insertion(+), 2 deletions(-)

diff --git a/build_tools/azure/python_nogil_lock.txt b/build_tools/azure/python_nogil_lock.txt
index 9dba84dfb619e..85988263b94b4 100644
--- a/build_tools/azure/python_nogil_lock.txt
+++ b/build_tools/azure/python_nogil_lock.txt
@@ -55,7 +55,7 @@ scipy==1.9.3
     # via -r /scikit-learn/build_tools/azure/python_nogil_requirements.txt
 six==1.16.0
     # via python-dateutil
-threadpoolctl==3.4.0
+threadpoolctl==3.3.0
     # via -r /scikit-learn/build_tools/azure/python_nogil_requirements.txt
 tomli==2.0.1
     # via pytest
diff --git a/build_tools/azure/python_nogil_requirements.txt b/build_tools/azure/python_nogil_requirements.txt
index 959b5ace581a8..970059ede81aa 100644
--- a/build_tools/azure/python_nogil_requirements.txt
+++ b/build_tools/azure/python_nogil_requirements.txt
@@ -13,4 +13,3 @@ joblib
 threadpoolctl
 pytest
 pytest-xdist
-meson-python