Port KMeans to use ProxyBase/InteropMixin

python/cuml/cuml/accel/_wrappers/sklearn/cluster.py

@@ -15,13 +15,22 @@
 #
 
 import cuml.cluster
+from cuml.accel.estimator_proxy import ProxyBase
 from cuml.accel.estimator_proxy_mixin import ProxyMixin
 
 __all__ = ("KMeans", "DBSCAN")
 
 
-class KMeans(ProxyMixin, cuml.cluster.KMeans):
-    pass
+class KMeans(ProxyBase):
+    _gpu_class = cuml.cluster.KMeans
+
+    def _gpu_fit_transform(self, X, y=None, sample_weight=None):
+        # Fixes signature mismatch with cuml.KMeans. Can be removed after #6741.
+        return self._gpu.fit_transform(X, y=y, sample_weight=sample_weight)
+
+    def _init_centroids(self, *args, **kwargs):
+        # Exposed for use by the sklearn test suite
+        return self._cpu._init_centroids(*args, **kwargs)
 
 
 class DBSCAN(ProxyMixin, cuml.cluster.DBSCAN):

python/cuml/cuml/accel/estimator_proxy.py

@@ -135,6 +135,11 @@ def __init__(self, *args, **kwargs):
         # be pickled properly.
         cls.__module__ = cls._gpu_class._cpu_class_path.rsplit(".", 1)[0]
 
+        # Forward _estimator_type as a class attribute if available
+        _estimator_type = getattr(cls._cpu_class, "_estimator_type", None)
+        if isinstance(_estimator_type, str):
+            cls._estimator_type = _estimator_type
+
         # Add proxy method definitions for all public methods on CPU class
         # that aren't already defined on the proxy class
         methods = [

python/cuml/cuml/accel/tests/scikit-learn/xfail-list.yaml

@@ -76,17 +76,10 @@
   - "sklearn.semi_supervised.tests.test_self_training::test_classification[k_best-estimator0]"
   - "sklearn.semi_supervised.tests.test_self_training::test_classification[threshold-estimator0]"
   - "sklearn.semi_supervised.tests.test_self_training::test_zero_iterations[y1-estimator0]"
-  - "sklearn.tests.test_common::test_estimators[ElasticNet()-check_n_features_in_after_fitting]"
-  - "sklearn.tests.test_common::test_estimators[Lasso()-check_n_features_in_after_fitting]"
-  - "sklearn.tests.test_common::test_estimators[PCA()-check_n_features_in_after_fitting]"
-  - "sklearn.tests.test_common::test_estimators[TruncatedSVD()-check_n_features_in_after_fitting]"
-  - "sklearn.tests.test_common::test_estimators[LinearRegression()-check_n_features_in_after_fitting]"
   - "sklearn.tests.test_common::test_estimators[LogisticRegression()-check_estimator_sparse_tag]"
-  - "sklearn.tests.test_common::test_estimators[LogisticRegression()-check_n_features_in_after_fitting]"
   - "sklearn.tests.test_common::test_estimators[LogisticRegression()-check_non_transformer_estimators_n_iter]"
   - "sklearn.tests.test_common::test_estimators[LogisticRegression()-check_sample_weight_equivalence_on_dense_data]"
   - "sklearn.tests.test_common::test_estimators[LogisticRegression()-check_sample_weight_equivalence_on_sparse_data]"
-  - "sklearn.tests.test_common::test_estimators[Ridge()-check_n_features_in_after_fitting]"
   - "sklearn.tests.test_common::test_estimators[Ridge()-check_non_transformer_estimators_n_iter]"
   - "sklearn.utils.tests.test_validation::test_cross_val_predict[coo_matrix]"
 - reason: Test should fail with cuml.accel (scikit-learn 1.7)
@@ -696,7 +689,6 @@
   - "sklearn.linear_model.tests.test_sparse_coordinate_descent::test_sparse_dense_equality[csc_matrix-True-6-24-True-Lasso]"
   - "sklearn.linear_model.tests.test_sparse_coordinate_descent::test_sparse_lasso_not_as_toy_dataset[csc_array]"
   - "sklearn.linear_model.tests.test_sparse_coordinate_descent::test_sparse_lasso_not_as_toy_dataset[csc_matrix]"
-  - "sklearn.manifold.tests.test_spectral_embedding::test_pipeline_spectral_clustering"
   - "sklearn.manifold.tests.test_t_sne::test_accessible_kl_divergence"
   - "sklearn.manifold.tests.test_t_sne::test_bad_precomputed_distances[D0-.* square distance matrix-barnes_hut-asarray]"
   - "sklearn.manifold.tests.test_t_sne::test_bad_precomputed_distances[D0-.* square distance matrix-barnes_hut-csr_array]"
@@ -1226,6 +1218,8 @@
   - "sklearn.manifold.tests.test_t_sne::test_optimization_minimizes_kl_divergence"
   - "sklearn.model_selection.tests.test_validation::test_cross_val_predict[coo_array]"
   - "sklearn.model_selection.tests.test_validation::test_cross_val_predict[coo_matrix]"
+  - "sklearn.manifold.tests.test_t_sne::test_uniform_grid[barnes_hit]"
+  - "sklearn.manifold.tests.test_spectral_embedding::test_pipeline_spectral_clustering"
 - reason: "cuml.accel bug: Missing components_ attribute"
   tests:
   - "sklearn.cluster.tests.test_dbscan::test_dbscan_no_core_samples[csr_array]"
@@ -1262,9 +1256,6 @@
 - reason: "cuml.accel does not support callable initialization for KMeans"
   marker: cuml_accel_kmeans_callable_init
   tests:
-  - "sklearn.cluster.tests.test_k_means::test_all_init[KMeans-callable-dense]"
-  - "sklearn.cluster.tests.test_k_means::test_all_init[KMeans-callable-sparse_array]"
-  - "sklearn.cluster.tests.test_k_means::test_all_init[KMeans-callable-sparse_matrix]"
   - "sklearn.cluster.tests.test_k_means::test_kmeans_init_auto_with_initial_centroids[KMeans-<lambda>-default]"
 - reason: "cuml.accel fails for KMeans with AttributeError: 'KMeans' object has no attribute '_n_init'"
   marker: cuml_accel_kmeans_n_init
@@ -1306,13 +1297,6 @@
   - "sklearn.cluster.tests.test_k_means::test_dense_sparse[42-KMeans-X_csr0]"
   - "sklearn.cluster.tests.test_k_means::test_dense_sparse[42-KMeans-X_csr1]"
   - "sklearn.cluster.tests.test_spectral::test_precomputed_nearest_neighbors_filtering"
-- reason: "cuml.accel fails on sparse matrix inputs for KMeans"
-  marker: cuml_accel_kmeans_sparse_unsupported
-  tests:
-  - "sklearn.cluster.tests.test_k_means::test_predict_dense_sparse[KMeans-k-means++-X_csr0]"
-  - "sklearn.cluster.tests.test_k_means::test_predict_dense_sparse[KMeans-k-means++-X_csr1]"
-  - "sklearn.cluster.tests.test_k_means::test_predict_dense_sparse[KMeans-random-X_csr0]"
-  - "sklearn.cluster.tests.test_k_means::test_predict_dense_sparse[KMeans-random-X_csr1]"
 - reason: "Integer inputs are converted to float32 instead of float64 with cuml.accel"
   marker: cuml_accel_kmeans_integer_dtype
   tests:
@@ -1329,19 +1313,13 @@
   tests:
   - "sklearn.cluster.tests.test_k_means::test_wrong_params[param0-n_samples.* should be >= n_clusters-KMeans]"
   - "sklearn.cluster.tests.test_k_means::test_wrong_params[param1-The shape of the initial centers .* does not match the number of clusters-KMeans]"
-  - "sklearn.cluster.tests.test_k_means::test_wrong_params[param2-The shape of the initial centers .* does not match the number of clusters-KMeans]"
   - "sklearn.cluster.tests.test_k_means::test_wrong_params[param3-The shape of the initial centers .* does not match the number of features of the data-KMeans]"
-  - "sklearn.cluster.tests.test_k_means::test_wrong_params[param4-The shape of the initial centers .* does not match the number of features of the data-KMeans]"
   - "sklearn.cluster.tests.test_k_means::test_relocating_with_duplicates[elkan-dense]"
   - "sklearn.cluster.tests.test_k_means::test_relocating_with_duplicates[lloyd-dense]"
 - reason: "cuml.accel deviates in KMeans empty cluster reloacation (number of labels)"
   marker: cuml_accel_kmeans_empty_clusters_num_labels
   tests:
   - "sklearn.cluster.tests.test_k_means::test_kmeans_empty_cluster_relocated[dense]"
-- reason: "cuml.accel fails with AttributeError: 'memmap' object has no attribute 'ptr'."
-  marker: cuml-accel_kmeans_fails_with_attribute_error_memmap_ptr
-  tests:
-  - "sklearn.cluster.tests.test_k_means::test_predict_does_not_change_cluster_centers[None]"
 - reason: "The signs of the principal components are swapped with cuml.accel (insignificant deviation)"
   condition: "cuda-python >= 12.9"
   tests:
@@ -1437,6 +1415,16 @@
   - "sklearn.tests.test_common::test_estimators[TSNE()-check_estimators_empty_data_messages]"
   - "sklearn.tests.test_common::test_estimators[TSNE()-check_estimators_nan_inf]"
   - "sklearn.tests.test_common::test_estimators[TSNE()-check_estimator_sparse_matrix]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_sample_weights_not_an_array]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_sample_weights_shape]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_sample_weight_equivalence_on_dense_data]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_sample_weight_equivalence_on_sparse_data]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_complex_data]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_dtype_object]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_estimators_nan_inf]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_transformer_data_not_an_array]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_fit1d]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_fit2d_predict1d]"
 - reason: "cuml doesn't set `feature_names_in_` properly"
   marker: cuml_accel_feature_names_in
   tests:
@@ -1448,16 +1436,26 @@
   - "sklearn.tests.test_common::test_pandas_column_name_consistency[PCA()]"
   - "sklearn.tests.test_common::test_pandas_column_name_consistency[TruncatedSVD()]"
   - "sklearn.tests.test_common::test_pandas_column_name_consistency[TSNE()]"
+  - "sklearn.tests.test_common::test_pandas_column_name_consistency[KMeans()]"
 - reason: "cuml raises a different error if X doesn't have expected n features"
   marker: cuml_accel_check_n_features_in
   tests:
   - "sklearn.tests.test_common::test_check_n_features_in_after_fitting[ElasticNet()]"
+  - "sklearn.tests.test_common::test_check_n_features_in_after_fitting[KMeans()]"
   - "sklearn.tests.test_common::test_check_n_features_in_after_fitting[Lasso()]"
   - "sklearn.tests.test_common::test_check_n_features_in_after_fitting[LinearRegression()]"
   - "sklearn.tests.test_common::test_check_n_features_in_after_fitting[LogisticRegression()]"
-  - "sklearn.tests.test_common::test_check_n_features_in_after_fitting[Ridge()]"
   - "sklearn.tests.test_common::test_check_n_features_in_after_fitting[PCA()]"
+  - "sklearn.tests.test_common::test_check_n_features_in_after_fitting[Ridge()]"
   - "sklearn.tests.test_common::test_check_n_features_in_after_fitting[TruncatedSVD()]"
+  - "sklearn.tests.test_common::test_estimators[ElasticNet()-check_n_features_in_after_fitting]"
+  - "sklearn.tests.test_common::test_estimators[KMeans()-check_n_features_in_after_fitting]"
+  - "sklearn.tests.test_common::test_estimators[Lasso()-check_n_features_in_after_fitting]"
+  - "sklearn.tests.test_common::test_estimators[LinearRegression()-check_n_features_in_after_fitting]"
+  - "sklearn.tests.test_common::test_estimators[LogisticRegression()-check_n_features_in_after_fitting]"
+  - "sklearn.tests.test_common::test_estimators[PCA()-check_n_features_in_after_fitting]"
+  - "sklearn.tests.test_common::test_estimators[Ridge()-check_n_features_in_after_fitting]"
+  - "sklearn.tests.test_common::test_estimators[TruncatedSVD()-check_n_features_in_after_fitting]"
 - reason: "cuml missing certain fit attributes"
   marker: cuml_accel_missing_fit_attributes
   tests: