FIX Allow for KMean's attributes to be readonly (scikit-learn#24258)

thomasjpfan · jjerphan · glemaitre · commit b402b4a67869 · 2022-09-12T19:53:59.000+02:00
Co-authored-by: Julien Jerphanion &lt;git@jjerphan.xyz&gt;
diff --git a/doc/whats_new/v1.2.rst b/doc/whats_new/v1.2.rst
@@ -138,6 +138,9 @@ Changelog
   :class:`cluster.AgglomerativeClustering` and will be renamed to `metric` in v1.4.
   :pr:`23470` by :user:`Meekail Zain <micky774>`.
 
+- |Fix| :class:`cluster.KMeans` now supports readonly attributes when predicting.
+  :pr:`24258` by `Thomas Fan`_
+
 :mod:`sklearn.datasets`
 .......................
 
diff --git a/sklearn/cluster/_k_means_lloyd.pyx b/sklearn/cluster/_k_means_lloyd.pyx
@@ -54,14 +54,13 @@ def lloyd_iter_chunked_dense(
 
     centers_new : ndarray of shape (n_clusters, n_features), dtype=floating
         Centers after previous iteration, placeholder for the new centers
-        computed during this iteration.
-
-    centers_squared_norms : ndarray of shape (n_clusters,), dtype=floating
-        Squared L2 norm of the centers.
+        computed during this iteration. `centers_new` can be `None` if
+        `update_centers` is False.
 
     weight_in_clusters : ndarray of shape (n_clusters,), dtype=floating
         Placeholder for the sums of the weights of every observation assigned
-        to each center.
+        to each center. `weight_in_clusters` can be `None` if `update_centers`
+        is False.
 
     labels : ndarray of shape (n_samples,), dtype=int
         labels assignment.
@@ -82,7 +81,7 @@ def lloyd_iter_chunked_dense(
     cdef:
         int n_samples = X.shape[0]
         int n_features = X.shape[1]
-        int n_clusters = centers_new.shape[0]
+        int n_clusters = centers_old.shape[0]
 
         # hard-coded number of samples per chunk. Appeared to be close to
         # optimal in all situations.
@@ -253,14 +252,13 @@ def lloyd_iter_chunked_sparse(
 
     centers_new : ndarray of shape (n_clusters, n_features), dtype=floating
         Centers after previous iteration, placeholder for the new centers
-        computed during this iteration.
-
-    centers_squared_norms : ndarray of shape (n_clusters,), dtype=floating
-        Squared L2 norm of the centers.
+        computed during this iteration. `centers_new` can be `None` if
+        `update_centers` is False.
 
     weight_in_clusters : ndarray of shape (n_clusters,), dtype=floating
         Placeholder for the sums of the weights of every observation assigned
-        to each center.
+        to each center. `weight_in_clusters` can be `None` if `update_centers`
+        is False.
 
     labels : ndarray of shape (n_samples,), dtype=int
         labels assignment.
@@ -282,7 +280,7 @@ def lloyd_iter_chunked_sparse(
     cdef:
         int n_samples = X.shape[0]
         int n_features = X.shape[1]
-        int n_clusters = centers_new.shape[0]
+        int n_clusters = centers_old.shape[0]
 
         # Choose same as for dense. Does not have the same impact since with
         # sparse data the pairwise distances matrix is not precomputed.
diff --git a/sklearn/cluster/_kmeans.py b/sklearn/cluster/_kmeans.py
@@ -784,8 +784,7 @@ def _labels_inertia(
     n_clusters = centers.shape[0]
 
     labels = np.full(n_samples, -1, dtype=np.int32)
-    weight_in_clusters = np.zeros(n_clusters, dtype=centers.dtype)
-    center_shift = np.zeros_like(weight_in_clusters)
+    center_shift = np.zeros(n_clusters, dtype=centers.dtype)
 
     if sp.issparse(X):
         _labels = lloyd_iter_chunked_sparse
@@ -795,16 +794,17 @@ def _labels_inertia(
         _inertia = _inertia_dense
         X = ReadonlyArrayWrapper(X)
 
+    centers = ReadonlyArrayWrapper(centers)
     _labels(
         X,
         sample_weight,
         x_squared_norms,
         centers,
-        centers,
-        weight_in_clusters,
-        labels,
-        center_shift,
-        n_threads,
+        centers_new=None,
+        weight_in_clusters=None,
+        labels=labels,
+        center_shift=center_shift,
+        n_threads=n_threads,
         update_centers=False,
     )
 
diff --git a/sklearn/cluster/tests/test_k_means.py b/sklearn/cluster/tests/test_k_means.py
@@ -29,6 +29,7 @@
 from sklearn.cluster._k_means_common import _inertia_dense
 from sklearn.cluster._k_means_common import _inertia_sparse
 from sklearn.cluster._k_means_common import _is_same_clustering
+from sklearn.utils._testing import create_memmap_backed_data
 from sklearn.datasets import make_blobs
 from io import StringIO
 
@@ -1213,3 +1214,23 @@ def test_feature_names_out(Klass, method):
 
     names_out = kmeans.get_feature_names_out()
     assert_array_equal([f"{class_name}{i}" for i in range(n_clusters)], names_out)
+
+
+@pytest.mark.parametrize("is_sparse", [True, False])
+def test_predict_does_not_change_cluster_centers(is_sparse):
+    """Check that predict does not change cluster centers.
+
+    Non-regression test for gh-24253.
+    """
+    X, _ = make_blobs(n_samples=200, n_features=10, centers=10, random_state=0)
+    if is_sparse:
+        X = sp.csr_matrix(X)
+
+    kmeans = KMeans()
+    y_pred1 = kmeans.fit_predict(X)
+    # Make cluster_centers readonly
+    kmeans.cluster_centers_ = create_memmap_backed_data(kmeans.cluster_centers_)
+    kmeans.labels_ = create_memmap_backed_data(kmeans.labels_)
+
+    y_pred2 = kmeans.predict(X)
+    assert_array_equal(y_pred1, y_pred2)
