test_svm fails on i386 with scipy 1.13 #29633
Description
Describe the bug
scipy 1.13 is triggering test failure in test_svc_ovr_tie_breaking[NuSVC] on i386 architecture.
The error can be seeing in debian CI tests, https://ci.debian.net/packages/s/scikit-learn/unstable/i386/
Full test log at https://ci.debian.net/packages/s/scikit-learn/unstable/i386/50043538/
or https://ci.debian.net/packages/s/scikit-learn/testing/i386/50043537/
Steps/Code to Reproduce
On an i386 system with scipy 1.13 installed
$ pytest-3 /usr/lib/python3/dist-packages/sklearn/svm/tests/test_svm.py -k test_svc_ovr_tie_breaking
Expected Results
test should pass
Actual Results
1333s _______________________ test_svc_ovr_tie_breaking[NuSVC] _______________________
1333s
1333s SVCClass = <class 'sklearn.svm._classes.NuSVC'>
1333s
1333s @pytest.mark.parametrize("SVCClass", [svm.SVC, svm.NuSVC])
1333s def test_svc_ovr_tie_breaking(SVCClass):
1333s """Test if predict breaks ties in OVR mode.
1333s Related issue: https://github.com/scikit-learn/scikit-learn/issues/8277
1333s """
1333s X, y = make_blobs(random_state=0, n_samples=20, n_features=2)
1333s
1333s xs = np.linspace(X[:, 0].min(), X[:, 0].max(), 100)
1333s ys = np.linspace(X[:, 1].min(), X[:, 1].max(), 100)
1333s xx, yy = np.meshgrid(xs, ys)
1333s
1333s common_params = dict(
1333s kernel="rbf", gamma=1e6, random_state=42, decision_function_shape="ovr"
1333s )
1333s svm = SVCClass(
1333s break_ties=False,
1333s **common_params,
1333s ).fit(X, y)
1333s pred = svm.predict(np.c_[xx.ravel(), yy.ravel()])
1333s dv = svm.decision_function(np.c_[xx.ravel(), yy.ravel()])
1333s > assert not np.all(pred == np.argmax(dv, axis=1))
1333s E assert not True
1333s E + where True = <function all at 0xf689d5e0>(array([1, 1, 1, ..., 1, 1, 1]) == array([1, 1, ..., dtype=int32)
1333s E + where <function all at 0xf689d5e0> = np.all
1333s E
1333s E Full diff:
1333s E - array([1, 1, 1, ..., 1, 1, 1], dtype=int32)
1333s E ? -------------
1333s E + array([1, 1, 1, ..., 1, 1, 1]))
1333s
1333s /usr/lib/python3/dist-packages/sklearn/svm/tests/test_svm.py:1225: AssertionError
Versions
$ python3 -c "import sklearn; sklearn.show_versions()"
# (on amd64, edited manually for i386)
System:
python: 3.12.4 (main, Jul 15 2024, 12:17:32) [GCC 13.3.0]
executable: /usr/bin/python3
machine: Linux-6.9.11-i386-i686-with-glibc2.39
Python dependencies:
sklearn: 1.4.2
pip: 24.1.1
setuptools: 70.3.0
numpy: 1.26.4
scipy: 1.13.1
Cython: 3.0.10
pandas: 2.2.2+dfsg
matplotlib: 3.8.3
joblib: 1.3.2
threadpoolctl: 3.1.0
Built with OpenMP: True
threadpoolctl info:
user_api: blas
internal_api: openblas
prefix: libopenblas
filepath: /usr/lib/i386-linux-gnu/openblas-pthread/libopenblasp-r0.3.27.so
version: 0.3.27
threading_layer: pthreads
architecture: Haswell
num_threads: 8
user_api: openmp
internal_api: openmp
prefix: libgomp
filepath: /usr/lib/i386-linux-gnu/libgomp.so.1.0.0
version: None