8000 MNT clean-up deprecations for 1.7: multi_class in LogisticRegression … · scikit-learn-bot/scikit-learn@f0c80e8 · GitHub
[go: up one dir, main page]
Skip to content

Commit f0c80e8

Browse files
jeremiedbbjeremiedbb
authored
MNT clean-up deprecations for 1.7: multi_class in LogisticRegression (scikit-learn#31241)
1 parent b985df0 commit f0c80e8

File tree

9 files changed

+112
-77
lines changed

9 files changed

+112
-77
lines changed

‎doc/modules/model_evaluation.rst

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1632,7 +1632,7 @@ Therefore, the `y_score` parameter is of size (n_samples,).
16321632
>>> from sklearn.linear_model import LogisticRegression
16331633
>>> from sklearn.metrics import roc_auc_score
16341634
>>> X, y = load_breast_cancer(return_X_y=True)
1635-
>>> clf = LogisticRegression(solver="liblinear").fit(X, y)
1635+
>>> clf = LogisticRegression().fit(X, y)
16361636
>>> clf.classes_
16371637
array([0, 1])
16381638

@@ -1728,11 +1728,11 @@ class with the greater label for each output.
17281728
>>> from sklearn.datasets import make_multilabel_classification
17291729
>>> from sklearn.multioutput import MultiOutputClassifier
17301730
>>> X, y = make_multilabel_classification(random_state=0)
1731-
>>> inner_clf = LogisticRegression(solver="liblinear", random_state=0)
1731+
>>> inner_clf = LogisticRegression(random_state=0)
17321732
>>> clf = MultiOutputClassifier(inner_clf).fit(X, y)
17331733
>>> y_score = np.transpose([y_pred[:, 1] for y_pred in clf.predict_proba(X)])
17341734
>>> roc_auc_score(y, y_score, average=None)
1735-
array([0.82..., 0.86..., 0.94..., 0 6D40 .85... , 0.94...])
1735+
array([0.82..., 0.85..., 0.93..., 0.86..., 0.94...])
17361736

17371737
And the decision values do not require such processing.
17381738

‎doc/whats_new/upcoming_changes/sklearn.linear_model/31241.api.rst

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,7 @@
1+
- Using the `"liblinear"` solver for multiclass classification with a one-versus-rest
2+
scheme in :class:`linear_model.LogisticRegression` and
3+
:class:`linear_model.LogisticRegressionCV` is deprecated and will raise an error in
4+
version 1.8. Either use a solver which supports the multinomial loss or wrap the
5+
estimator in a :class:`sklearn.multiclass.OneVsRestClassifier` to keep applying a
6+
one-versus-rest scheme.
7+
By :user:`Jérémie du Boisberranger <jeremiedbb>`.

‎sklearn/ensemble/tests/test_voting.py

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -114,7 +114,7 @@ def test_notfitted():
114114

115115
def test_majority_label_iris(global_random_seed):
116116
"""Check classification by majority label on dataset iris."""
117-
clf1 = LogisticRegression(solver="liblinear", random_state=global_random_seed)
117+
clf1 = LogisticRegression(random_state=global_random_seed)
118118
clf2 = RandomForestClassifier(n_estimators=10, random_state=global_random_seed)
119119
clf3 = GaussianNB()
120120
eclf = VotingClassifier(
@@ -127,12 +127,12 @@ def test_majority_label_iris(global_random_seed):
127127

128128
def test_tie_situation():
129129
"""Check voting classifier selects smaller class label in tie situation."""
130-
clf1 = LogisticRegression(random_state=123, solver="liblinear")
130+
clf1 = LogisticRegression(random_state=123)
131131
clf2 = RandomForestClassifier(random_state=123)
132132
eclf = VotingClassifier(estimators=[("lr", clf1), ("rf", clf2)], voting="hard")
133-
assert clf1.fit(X, y).predict(X)[73] == 2
134-
assert clf2.fit(X, y).predict(X)[73] == 1
135-
assert eclf.fit(X, y).predict(X)[73] == 1
133+
assert clf1.fit(X, y).predict(X)[52] == 2
134+
assert clf2.fit(X, y).predict(X)[52] == 1
135+
assert eclf.fit(X, y).predict(X)[52] == 1
136136

137137

138138
def test_weights_iris(global_random_seed):

‎sklearn/linear_model/_logistic.py

Lines changed: 22 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -501,6 +501,15 @@ def _logistic_regression_path(
501501
w0 = sol.solve(X=X, y=target, sample_weight=sample_weight)
502502
n_iter_i = sol.iteration
503503
elif solver == "liblinear":
504+
if len(classes) > 2:
505+
warnings.warn(
506+
"Using the 'liblinear' solver for multiclass classification is "
507+
"deprecated. An error will be raised in 1.8. Either use another "
508+
"solver which supports the multinomial loss or wrap the estimator "
509+
"in a OneVsRestClassifier to keep applying a one-versus-rest "
510+
"scheme.",
511+
FutureWarning,
512+
)
504513
(
505514
coef_,
506515
intercept_,
@@ -931,7 +940,7 @@ class LogisticRegression(LinearClassifierMixin, SparseCoefMixin, BaseEstimator):
931940
'lbfgs' 'l2', None yes
932941
'liblinear' 'l1', 'l2' no
933942
'newton-cg' 'l2', None yes
934-
'newton-cholesky' 'l2', None no
943+
'newton-cholesky' 'l2', None yes
935944
'sag' 'l2', None yes
936945
'saga' 'elasticnet', 'l1', 'l2', None yes
937946
================= ============================== ======================
@@ -1238,7 +1247,7 @@ def fit(self, X, y, sample_weight=None):
12381247
check_classification_targets(y)
12391248
self.classes_ = np.unique(y)
12401249

1241-
# TODO(1.7) remove multi_class
1250+
# TODO(1.8) remove multi_class
12421251
multi_class = self.multi_class
12431252
if self.multi_class == "multinomial" and len(self.classes_) == 2:
12441253
warnings.warn(
@@ -1274,6 +1283,15 @@ def fit(self, X, y, sample_weight=None):
12741283
multi_class = _check_multi_class(multi_class, solver, len(self.classes_))
12751284

12761285
if solver == "liblinear":
1286+
if len(self.classes_) > 2:
1287+
warnings.warn(
1288+
"Using the 'liblinear' solver for multiclass classification is "
1289+
"deprecated. An error will be raised in 1.8. Either use another "
1290+
"solver which supports the multinomial loss or wrap the estimator "
1291+
"in a OneVsRestClassifier to keep applying a one-versus-rest "
1292+
"scheme.",
1293+
FutureWarning,
1294+
)
12771295
if effective_n_jobs(self.n_jobs) != 1:
12781296
warnings.warn(
12791297
"'n_jobs' > 1 does not have any effect when"
@@ -1568,7 +1586,7 @@ class LogisticRegressionCV(LogisticRegression, LinearClassifierMixin, BaseEstima
15681586
'lbfgs' 'l2' yes
15691587
'liblinear' 'l1', 'l2' no
15701588
'newton-cg' 'l2' yes
1571-
'newton-cholesky' 'l2', no
1589+
'newton-cholesky' 'l2', yes
15721590
'sag' 'l2', yes
15731591
'saga' 'elasticnet', 'l1', 'l2' yes
15741592
================= ============================== ======================
@@ -1900,7 +1918,7 @@ def fit(self, X, y, sample_weight=None, **params):
19001918
classes = self.classes_ = label_encoder.classes_
19011919
encoded_labels = label_encoder.transform(label_encoder.classes_)
19021920

1903-
# TODO(1.7) remove multi_class
1921+
# TODO(1.8) remove multi_class
19041922
multi_class = self.multi_class
19051923
if self.multi_class == "multinomial" and len(self.classes_) == 2:
19061924
warnings.warn(

‎sklearn/linear_model/tests/test_logistic.py

Lines changed: 42 additions & 19 deletions
Original file line numberDiff line numberDiff line change
@@ -129,8 +129,7 @@ def __call__(self, model, X, y, sample_weight=None):
129129

130130
@skip_if_no_parallel
131131
def test_lr_liblinear_warning():
132-
n_samples, n_features = iris.data.shape
133-
target = iris.target_names[iris.target]
132+
X, y = make_classification(random_state=0)
134133

135134
lr = LogisticRegression(solver="liblinear", n_jobs=2)
136135
warning_message = (
@@ -139,7 +138,7 @@ def test_lr_liblinear_warning():
139138
" = 2."
140139
)
141140
with pytest.warns(UserWarning, match=warning_message):
142-
lr.fit(iris.data, target)
141+
lr.fit(X, y)
143142

144143

145144
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
@@ -148,8 +147,11 @@ def test_predict_3_classes(csr_container):
148147
check_predictions(LogisticRegression(C=10), csr_container(X), Y2)
149148

150149

151-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
150+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
152151
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
152+
@pytest.mark.filterwarnings(
153+
"ignore:.*'liblinear' solver for multiclass classification is deprecated.*"
154+
)
153155
@pytest.mark.parametrize(
154156
"clf",
155157
[
@@ -197,7 +199,7 @@ def test_predict_iris(clf):
197199
assert np.mean(pred == target) > 0.95
198200

199201

200-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
202+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
201203
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
202204
@pytest.mark.parametrize("LR", [LogisticRegression, LogisticRegressionCV])
203205
def test_check_solver_option(LR):
@@ -249,7 +251,7 @@ def test_elasticnet_l1_ratio_err_helpful(LR):
249251
model.fit(np.array([[1, 2], [3, 4]]), np.array([0, 1]))
250252

251253

252-
# TODO(1.7): remove whole test with deprecation of multi_class
254+
# TODO(1.8): remove whole test with deprecation of multi_class
253255
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
254256
@pytest.mark.parametrize("solver", ["lbfgs", "newton-cg", "sag", "saga"])
255257
def test_multinomial_binary(solver):
@@ -274,7 +276,7 @@ def test_multinomial_binary(solver):
274276
assert np.mean(pred == target) > 0.9
275277

276278

277-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
279+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
278280
# Maybe even remove this whole test as correctness of multinomial loss is tested
279281
# elsewhere.
280282
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
@@ -614,7 +616,7 @@ def test_logistic_cv_sparse(csr_container):
614616
assert clfs.C_ == clf.C_
615617

616618

617-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
619+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
618620
# Best remove this whole test.
619621
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
620622
def test_ovr_multinomial_iris():
@@ -700,7 +702,7 @@ def test_logistic_regression_solvers():
700702
)
701703

702704

703-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
705+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
704706
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
705707
@pytest.mark.parametrize("fit_intercept", [False, True])
706708
def test_logistic_regression_solvers_multiclass(fit_intercept):
@@ -1301,7 +1303,7 @@ def test_logreg_predict_proba_multinomial():
13011303
assert clf_wrong_loss > clf_multi_loss
13021304

13031305

1304-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
1306+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
13051307
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
13061308
@pytest.mark.parametrize("max_iter", np.arange(1, 5))
13071309
@pytest.mark.parametrize("multi_class", ["ovr", "multinomial"])
@@ -1345,8 +1347,11 @@ def test_max_iter(max_iter, multi_class, solver, message):
13451347
assert lr.n_iter_[0] == max_iter
13461348

13471349

1348-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
1350+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
1349 10000 1351
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
1352+
@pytest.mark.filterwarnings(
1353+
"ignore:.*'liblinear' solver for multiclass classification is deprecated.*"
1354+
)
13501355
@pytest.mark.parametrize("solver", SOLVERS)
13511356
def test_n_iter(solver):
13521357
# Test that self.n_iter_ has the correct format.
@@ -1478,7 +1483,7 @@ def test_saga_vs_liblinear(csr_container):
14781483
assert_array_almost_equal(saga.coef_, liblinear.coef_, 3)
14791484

14801485

1481-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
1486+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
14821487
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
14831488
@pytest.mark.parametrize("multi_class", ["ovr", "multinomial"])
14841489
@pytest.mark.parametrize(
@@ -1738,7 +1743,7 @@ def test_LogisticRegressionCV_GridSearchCV_elastic_net(n_classes):
17381743
assert gs.best_params_["C"] == lrcv.C_[0]
17391744

17401745

1741-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
1746+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
17421747
# Maybe remove whole test after removal of the deprecated multi_class.
17431748
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
17441749
def test_LogisticRegressionCV_GridSearchCV_elastic_net_ovr():
@@ -1786,7 +1791,7 @@ def test_LogisticRegressionCV_GridSearchCV_elastic_net_ovr():
17861791
assert (lrcv.predict(X_test) == gs.predict(X_test)).mean() >= 0.8
17871792

17881793

1789-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
1794+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
17901795
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
17911796
@pytest.mark.parametrize("penalty", ("l2", "elasticnet"))
17921797
@pytest.mark.parametrize("multi_class", ("ovr", "multinomial", "auto"))
@@ -1825,7 +1830,7 @@ def test_LogisticRegressionCV_no_refit(penalty, multi_class):
18251830
assert lrcv.coef_.shape == (n_classes, n_features)
18261831

18271832

1828-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
1833+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
18291834
# Remove multi_class an change first element of the expected n_iter_.shape from
18301835
# n_classes to 1 (according to the docstring).
18311836
@pytest 10000 .mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
@@ -1955,8 +1960,11 @@ def test_logistic_regression_path_coefs_multinomial():
19551960
assert_array_almost_equal(coefs[1], coefs[2], decimal=1)
19561961

19571962

1958-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
1963+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
19591964
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
1965+
@pytest.mark.filterwarnings(
1966+
"ignore:.*'liblinear' solver for multiclass classification is deprecated.*"
1967+
)
19601968
@pytest.mark.parametrize(
19611969
"est",
19621970
[
@@ -2126,7 +2134,7 @@ def test_scores_attribute_layout_elasticnet():
21262134
assert avg_scores_lrcv[i, j] == pytest.approx(avg_score_lr)
21272135

21282136

2129-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
2137+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
21302138
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
21312139
@pytest.mark.parametrize("solver", ["lbfgs", "newton-cg", "newton-cholesky"])
21322140
@pytest.mark.parametrize("fit_intercept", [False, True])
@@ -2171,7 +2179,7 @@ def test_multinomial_identifiability_on_iris(solver, fit_intercept):
21712179
assert clf.intercept_.sum(axis=0) == pytest.approx(0, abs=1e-11)
21722180

21732181

2174-
# TODO(1.7): remove filterwarnings after the deprecation of multi_class
2182+
# TODO(1.8): remove filterwarnings after the deprecation of multi_class
21752183
@pytest.mark.filterwarnings("ignore:.*'multi_class' was deprecated.*:FutureWarning")
21762184
@pytest.mark.parametrize("multi_class", ["ovr", "multinomial", "auto"])
21772185
@pytest.mark.parametrize("class_weight", [{0: 1.0, 1: 10.0, 2: 1.0}, "balanced"])
@@ -2349,7 +2357,7 @@ def test_passing_params_without_enabling_metadata_routing():
23492357
lr_cv.score(X, y, **params)
23502358

23512359

2352-
# TODO(1.7): remove
2360+
# TODO(1.8): remove
23532361
def test_multi_class_deprecated():
23542362
"""Check `multi_class` parameter deprecated."""
23552363
X, y = make_classification(n_classes=3, n_samples=50, n_informative=6)
@@ -2414,3 +2422,18 @@ def test_newton_cholesky_fallback_to_lbfgs(global_random_seed):
24142422
n_iter_nc_limited = lr_nc_limited.n_iter_[0]
24152423

24162424
assert n_iter_nc_limited == lr_nc_limited.max_iter - 1
2425+
2426+
2427+
# TODO(1.8): check for an error instead
2428+
@pytest.mark.parametrize("Estimator", [LogisticRegression, LogisticRegressionCV])
2429+
def test_liblinear_multiclass_warning(Estimator):
2430+
"""Check that liblinear warns on multiclass problems."""
2431+
msg = (
2432+
"Using the 'liblinear' solver for multiclass classification is "
2433+
"deprecated. An error will be raised in 1.8. Either use another "
2434+
"solver which supports the multinomial loss or wrap the estimator "
2435+
"in a OneVsRestClassifier to keep applying a one-versus-rest "
2436+
"scheme."
2437+
)
2438+
with pytest.warns(FutureWarning, match=msg):
2439+
Estimator(solver="liblinear").fit(iris.data, iris.target)

‎sklearn/metrics/_ranking.py

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -622,7 +622,7 @@ class scores must correspond to the order of ``labels``,
622622
>>> from sklearn.linear_model import LogisticRegression
623623
>>> from sklearn.metrics import roc_auc_score
624624
>>> X, y = load_breast_cancer(return_X_y=True)
625-
>>> clf = LogisticRegression(solver="liblinear", random_state=0).fit(X, y)
625+
>>> clf = LogisticRegression(solver="newton-cholesky", random_state=0).fit(X, y)
626626
>>> roc_auc_score(y, clf.predict_proba(X)[:, 1])
627627
0.99...
628628
>>> roc_auc_score(y, clf.decision_function(X))
@@ -632,7 +632,7 @@ class scores must correspond to the order of ``labels``,
632632
633633
>>> from sklearn.datasets import load_iris
634634
>>> X, y = load_iris(return_X_y=True)
635-
>>> clf = LogisticRegression(solver="liblinear").fit(X, y)
635+
>>> clf = LogisticRegression(solver="newton-cholesky").fit(X, y)
636636
>>> roc_auc_score(y, clf.predict_proba(X), multi_class='ovr')
637637
0.99...
638638
@@ -649,7 +649,7 @@ class scores must correspond to the order of ``labels``,
649649
>>> # extract the positive columns for each output
650650
>>> y_score = np.transpose([score[:, 1] for score in y_score])
651651
>>> roc_auc_score(y, y_score, average=None)
652-
array([0.82..., 0.86..., 0.94..., 0.85... , 0.94...])
652+
array([0.82..., 0.85..., 0.93..., 0.86..., 0.94...])
653653
>>> from sklearn.linear_model import RidgeClassifierCV
654654
>>> clf = RidgeClassifierCV().fit(X, y)
655655
>>> roc_auc_score(y, clf.decision_function(X), average=None)

0 commit comments

Comments
 (0)
0