KMeans OOP

.ci/pipeline/build-and-test-lnx.yml

@@ -24,7 +24,7 @@ steps:
     displayName: "System info"
   - script: |
       conda update -y -q conda
-      conda create -q -y -n CB -c conda-forge -c intel python=$(PYTHON_VERSION) intel::dal-devel mpich pyyaml "dpcpp-cpp-rt=2024.0.2"
+      conda create -q -y -n CB -c conda-forge -c intel python=$(PYTHON_VERSION) intel::dal-devel mpich pyyaml "dpcpp-cpp-rt=2024.1.0"
     displayName: "Conda create"
   - script: |
       . /usr/share/miniconda/etc/profile.d/conda.sh

.ci/scripts/install_dpcpp.sh

@@ -21,5 +21,5 @@ rm GPG-PUB-KEY-INTEL-SW-PRODUCTS-2023.PUB
 echo "deb https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list
 sudo add-apt-repository -y "deb https://apt.repos.intel.com/oneapi all main"
 sudo apt-get update
-sudo apt-get install -y intel-dpcpp-cpp-compiler-2024.0
+sudo apt-get install -y intel-dpcpp-cpp-compiler-2024.1
 sudo bash -c 'echo libintelocl.so > /etc/OpenCL/vendors/intel-cpu.icd'

deselected_tests.yaml

@@ -167,7 +167,7 @@ deselected_tests:
 
   # test_non_uniform_strategies fails due to differences in handling of vacuous clusters after update
   # See https://github.com/IntelPython/daal4py/issues/69
-  - cluster/tests/test_k_means.py::test_relocated_clusters >=0.23,<0.24
+  # - cluster/tests/test_k_means.py::test_relocated_clusters >=0.23,<0.24
   - cluster/tests/test_k_means.py::test_kmeans_relocated_clusters >=0.24
 
   # In scikit-learn, these algorithms are not included in this test. However, scikit-learn-intelex
@@ -178,6 +178,11 @@ deselected_tests:
   - cluster/tests/test_k_means.py::test_kmeans_convergence >=0.23
   - cluster/tests/test_k_means.py::test_kmeans_verbose >=0.23
 
+  # Sparse Support required
+  - cluster/tests/test_k_means.py::test_predict_sample_weight_deprecation_warning[KMeans] >=1.3
+  - cluster/tests/test_k_means.py::test_unit_weights_vs_no_weights[KMeans-dense] <1.2
+  - cluster/tests/test_k_means.py::test_unit_weights_vs_no_weights[42-KMeans-dense] >=1.2
+
   # The Newton-CG solver solution computed in float32 disagrees with that of float64 by a small
   # margin above the test threshold, see https://github.com/scikit-learn/scikit-learn/pull/13645
   - linear_model/tests/test_logistic.py::test_dtype_match
@@ -240,7 +245,7 @@ deselected_tests:
   - inspection/tests/test_permutation_importance.py::test_permutation_importance_sample_weight >=0.24
 
   # Patched and unpatched kmeans set same values to different clusters. Need to investigate.
-  - preprocessing/tests/test_discretization.py::test_nonuniform_strategies[kmeans-expected_2bins1-expected_3bins1-expected_5bins1] >=0.24
+  # - preprocessing/tests/test_discretization.py::test_nonuniform_strategies[kmeans-expected_2bins1-expected_3bins1-expected_5bins1] >=0.24
 
   # OOB scores in scikit-learn and oneDAL are different because of different random number generators
   - ensemble/tests/test_forest.py::test_forest_classifier_oob[X1-y1-0.65-array-ExtraTreesClassifier]  
@@ -346,12 +351,12 @@ deselected_tests:
   - tests/test_multioutput.py::test_classifier_chain_fit_and_predict_with_sparse_data >=1.4
 
   # New failing sklearn1.4.1 tests for kmeans associated with incorrect n_iter_ values in daal4py
-  - cluster/tests/test_k_means.py::test_relocating_with_duplicates[lloyd-dense] >=1.4
-  - cluster/tests/test_k_means.py::test_relocating_with_duplicates[lloyd-sparse_matrix] >=1.4
-  - cluster/tests/test_k_means.py::test_relocating_with_duplicates[lloyd-sparse_array] >=1.4
-  - cluster/tests/test_k_means.py::test_relocating_with_duplicates[elkan-dense] >=1.4
-  - cluster/tests/test_k_means.py::test_relocating_with_duplicates[elkan-sparse_matrix] >=1.4
-  - cluster/tests/test_k_means.py::test_relocating_with_duplicates[elkan-sparse_array] >=1.4
+  # - cluster/tests/test_k_means.py::test_relocating_with_duplicates[lloyd-dense] >=1.4
+  # - cluster/tests/test_k_means.py::test_relocating_with_duplicates[lloyd-sparse_matrix] >=1.4
+  # - cluster/tests/test_k_means.py::test_relocating_with_duplicates[lloyd-sparse_array] >=1.4
+  # - cluster/tests/test_k_means.py::test_relocating_with_duplicates[elkan-dense] >=1.4
+  # - cluster/tests/test_k_means.py::test_relocating_with_duplicates[elkan-sparse_matrix] >=1.4
+  # - cluster/tests/test_k_means.py::test_relocating_with_duplicates[elkan-sparse_array] >=1.4
 
 
   # --------------------------------------------------------
@@ -434,8 +439,8 @@ gpu:
 
   # Fails
   - cluster/tests/test_dbscan.py::test_weighted_dbscan
-  - cluster/tests/test_k_means.py::test_k_means_fit_predict
-  - cluster/tests/test_k_means.py::test_predict
+  # - cluster/tests/test_k_means.py::test_k_means_fit_predict
+  # - cluster/tests/test_k_means.py::test_predict
 
   - ensemble/tests/test_bagging.py::test_gridsearch
   - ensemble/tests/test_bagging.py::test_estimators_samples
@@ -567,8 +572,6 @@ gpu:
   - tests/test_common.py::test_estimators[GaussianMixture()-check_fit_idempotent]
   - tests/test_common.py::test_estimators[GaussianMixture()-check_n_features_in]
   - tests/test_common.py::test_estimators[GaussianMixture()-check_fit2d_predict1d]
-  - tests/test_common.py::test_estimators[KMeans()-check_clustering]
-  - tests/test_common.py::test_estimators[KMeans()-check_clustering(readonly_memmap=True)]
   - tests/test_common.py::test_estimators[RandomForestClassifier()-check_class_weight_classifiers]
   - tests/test_common.py::test_estimators[SVC()-check_sample_weights_pandas_series]
   - tests/test_common.py::test_estimators[SVC()-check_sample_weights_not_an_array]
@@ -603,7 +606,6 @@ gpu:
   - tests/test_multiclass.py::test_ovr_coef_
   - tests/test_multiclass.py::test_ovr_deprecated_coef_intercept
   - tests/test_multiclass.py::test_pairwise_cross_val_score
-
   - tests/test_multioutput.py::test_multiclass_multioutput_estimator_predict_proba
   - tests/test_multioutput.py::test_classifier_chain_fit_and_predict_with_sparse_data
 
@@ -617,21 +619,15 @@ gpu:
   - manifold/tests/test_t_sne.py::test_n_iter_without_progress
 
   # KMeans based (unsupported for GPU)
-  - cluster/tests/test_k_means.py
-  - tests/test_common.py::test_estimators[KMeans()
-  - tests/test_common.py::test_estimators[BayesianGaussianMixture()-check_fit_check_is_fitted]
-  - tests/test_common.py::test_estimators[GaussianMixture()-check_fit_check_is_fitted]
-  - tests/test_common.py::test_check_n_features_in_after_fitting[BayesianGaussianMixture()]
-  - tests/test_common.py::test_check_n_features_in_after_fitting[GaussianMixture()]
-  - tests/test_common.py::test_check_n_features_in_after_fitting[KMeans()]
-  - tests/test_common.py::test_set_output_transform[KMeans()]
-  - tests/test_common.py::test_set_output_transform_pandas[KMeans()]
-  - tests/test_common.py::test_global_output_transform_pandas[KMeans()]
-  - mixture/tests/test_gaussian_mixture.py
-  - model_selection/tests/test_validation.py::test_cross_val_predict
-  - metrics/tests/test_score_objects.py::test_supervised_cluster_scorers
-  - tests/test_pipeline.py::test_fit_predict_on_pipeline
-  - tests/test_discriminant_analysis.py::test_lda_predict
+  # - tests/test_common.py::test_estimators[BayesianGaussianMixture()-check_fit_check_is_fitted]
+  # - tests/test_common.py::test_estimators[GaussianMixture()-check_fit_check_is_fitted]
+  # - tests/test_common.py::test_check_n_features_in_after_fitting[BayesianGaussianMixture()]
+  # - tests/test_common.py::test_check_n_features_in_after_fitting[GaussianMixture()]
+  # - mixture/tests/test_gaussian_mixture.py
+  # - model_selection/tests/test_validation.py::test_cross_val_predict
+  # - metrics/tests/test_score_objects.py::test_supervised_cluster_scorers
+  # - tests/test_pipeline.py::test_fit_predict_on_pipeline
+  # - tests/test_discriminant_analysis.py::test_lda_predict
   # Other device issues
   - tests/test_metaestimators.py::test_meta_estimators_delegate_data_validation[StackingClassifier]
   - tests/test_multiclass.py::test_ovr_always_present
@@ -1105,9 +1101,3 @@ gpu:
   - tests/test_common.py::test_estimators[DBSCAN()-check_fit2d_predict1d]
   - tests/test_common.py::test_check_n_features_in_after_fitting[DBSCAN()]
   - tests/test_common.py::test_check_n_features_in_after_fitting[SVC()]
-
-preview:
-  - cluster/tests/test_k_means.py::test_kmeans_elkan_results
-  - cluster/tests/test_k_means.py::test_unit_weights_vs_no_weights[KMeans-dense] <1.2
-  - cluster/tests/test_k_means.py::test_unit_weights_vs_no_weights[42-KMeans-dense] >=1.2
-  - cluster/tests/test_k_means.py::test_predict_sample_weight_deprecation_warning[KMeans] >=1.3

onedal/cluster/kmeans.cpp

@@ -68,7 +68,12 @@ struct params2desc {
         desc.set_cluster_count( params["cluster_count"].cast<std::int64_t>() );
         desc.set_accuracy_threshold( params["accuracy_threshold"].cast<Float>() );
         desc.set_max_iteration_count( params["max_iteration_count"].cast<std::int64_t>() );
-
+#if defined(ONEDAL_VERSION) && ONEDAL_VERSION >= 20240200
+        auto result_options = params["result_options"].cast<std::string>();
+        if (result_options == "computeAssignments"){
+            desc.set_result_options(result_options::compute_assignments);
+        }
+#endif // defined(ONEDAL_VERSION) && ONEDAL_VERSION >= 20240200
         return desc;
     }
 };

onedal/cluster/kmeans.py

@@ -142,7 +142,7 @@ def _check_params_vs_input(
             self._n_init = 1
         assert self.algorithm == "lloyd"
 
-    def _get_onedal_params(self, dtype=np.float32):
+    def _get_onedal_params(self, dtype=np.float32, result_options=None):
         thr = self._tol if hasattr(self, "_tol") else self.tol
         return {
             "fptype": "float" if dtype == np.float32 else "double",
@@ -151,6 +151,7 @@ def _get_onedal_params(self, dtype=np.float32):
             "max_iteration_count": self.max_iter,
             "cluster_count": self.n_clusters,
             "accuracy_threshold": thr,
+            "result_options": "" if result_options is None else result_options,
         }
 
     def _get_params_and_input(self, X, policy):
@@ -340,7 +341,7 @@ def _set_cluster_centers(self, cluster_centers):
     cluster_centers_ = property(_get_cluster_centers, _set_cluster_centers)
 
     def _predict_raw(self, X_table, module, policy, dtype=np.float32):
-        params = self._get_onedal_params(dtype)
+        params = self._get_onedal_params(dtype, result_options="computeAssignments")
 
         result = module.infer(policy, params, self.model_, X_table)
 

setup_sklearnex.py

@@ -81,7 +81,6 @@
     "sklearnex.neighbors",
     "sklearnex.preview",
     "sklearnex.preview.covariance",
-    "sklearnex.preview.cluster",
     "sklearnex.svm",
     "sklearnex.utils",
 ]