MAINT CI cleanups and checks (scikit-learn#31690)

Author: ogrisel

.github/workflows/cuda-ci.yml

@@ -21,7 +21,7 @@ jobs:
         uses: pypa/cibuildwheel@5f22145df44122af0f5a201f93cf0207171beca7
         env:
           CIBW_BUILD: cp313-manylinux_x86_64
-          CIBW_MANYLINUX_X86_64_IMAGE: manylinux2014
+          CIBW_MANYLINUX_X86_64_IMAGE: manylinux_2_28
           CIBW_BUILD_VERBOSITY: 1
           CIBW_ARCHS: x86_64
 

build_tools/wheels/build_wheels.sh

@@ -49,13 +49,6 @@ if [[ $(uname) == "Darwin" ]]; then
     export LDFLAGS="$LDFLAGS -Wl,-rpath,$PREFIX/lib -L$PREFIX/lib -lomp"
 fi
 
-if [[ "$CIBW_FREE_THREADED_SUPPORT" =~ [tT]rue ]]; then
-    # Numpy, scipy, Cython only have free-threaded wheels on scientific-python-nightly-wheels
-    # TODO: remove this after CPython 3.13 is released (scheduled October 2024)
-    # and our dependencies have free-threaded wheels on PyPI
-    export CIBW_BUILD_FRONTEND='pip; args: --pre --extra-index-url "https://pypi.anaconda.org/scientific-python-nightly-wheels/simple" --only-binary :all:'
-fi
-
 # The version of the built dependencies are specified
 # in the pyproject.toml file, while the tests are run
 # against the most recent version of the dependencies

sklearn/svm/tests/test_sparse.py

@@ -125,6 +125,7 @@ def test_unsorted_indices(csr_container):
     X, y = load_digits(return_X_y=True)
     X_test = csr_container(X[50:100])
     X, y = X[:50], y[:50]
+    tols = dict(rtol=1e-12, atol=1e-14)
 
     X_sparse = csr_container(X)
     coef_dense = (
@@ -135,7 +136,7 @@ def test_unsorted_indices(csr_container):
     )
     coef_sorted = sparse_svc.coef_
     # make sure dense and sparse SVM give the same result
-    assert_allclose(coef_dense, coef_sorted.toarray())
+    assert_allclose(coef_dense, coef_sorted.toarray(), **tols)
 
     # reverse each row's indices
     def scramble_indices(X):
@@ -158,9 +159,11 @@ def scramble_indices(X):
     )
     coef_unsorted = unsorted_svc.coef_
     # make sure unsorted indices give same result
-    assert_allclose(coef_unsorted.toarray(), coef_sorted.toarray())
+    assert_allclose(coef_unsorted.toarray(), coef_sorted.toarray(), **tols)
     assert_allclose(
-        sparse_svc.predict_proba(X_test_unsorted), sparse_svc.predict_proba(X_test)
+        sparse_svc.predict_proba(X_test_unsorted),
+        sparse_svc.predict_proba(X_test),
+        **tols,
     )