FIX make sure that KernelPCA works with pandas output and arpack solver

doc/whats_new/v1.4.rst

@@ -253,6 +253,11 @@ Changelog
   :pr:`26315` and :pr:`27098` by :user:`Mateusz Sokół <mtsokol>`,
   :user:`Olivier Grisel <ogrisel>` and :user:`Edoardo Abati <EdAbati>`.
 
+- |Fix| Fixes a bug in :class:`decomposition.KernelPCA` by forcing the output of
+  the internal :class:`preprocessing.KernelCenterer` to be a default array. When the
+  arpack solver was used, it would expect an array with a `dtype` attribute.
+  :pr:`27583` by :user:`Guillaume Lemaitre <glemaitre>`.
+
 :mod:`sklearn.ensemble`
 .......................
 

sklearn/decomposition/_kernel_pca.py

@@ -432,7 +432,7 @@ def fit(self, X, y=None):
             raise ValueError("Cannot fit_inverse_transform with a precomputed kernel.")
         X = self._validate_data(X, accept_sparse="csr", copy=self.copy_X)
         self.gamma_ = 1 / X.shape[1] if self.gamma is None else self.gamma
-        self._centerer = KernelCenterer()
+        self._centerer = KernelCenterer().set_output(transform="default")
         K = self._get_kernel(X)
         self._fit_transform(K)
 

sklearn/decomposition/tests/test_kernel_pca.py

@@ -3,7 +3,8 @@
 import numpy as np
 import pytest
 
-from sklearn.datasets import make_blobs, make_circles
+import sklearn
+from sklearn.datasets import load_iris, make_blobs, make_circles
 from sklearn.decomposition import PCA, KernelPCA
 from sklearn.exceptions import NotFittedError
 from sklearn.linear_model import Perceptron
@@ -551,3 +552,15 @@ def test_kernel_pca_inverse_correct_gamma():
     X2_recon = kpca2.inverse_transform(kpca1.transform(X))
 
     assert_allclose(X1_recon, X2_recon)
+
+
+def test_kernel_pca_pandas_output():
+    """Check that KernelPCA works with pandas output when the solver is arpack.
+
+    Non-regression test for:
+    https://github.com/scikit-learn/scikit-learn/issues/27579
+    """
+    pytest.importorskip("pandas")
+    X, _ = load_iris(as_frame=True, return_X_y=True)
+    with sklearn.config_context(transform_output="pandas"):
+        KernelPCA(n_components=2, eigen_solver="arpack").fit_transform(X)