ENH: adding array-api-compat and enabling array api conformance tests (#2079)

* ENH: adding `array-api-compat` and enabling array api conformance tests

* Disabled tests and added comment

Author: samir-nasibli

deselected_tests.yaml

@@ -25,6 +25,26 @@
 #  will exclude deselection in versions 0.18.1, and 0.18.2 only.
 
 deselected_tests:
+  # Array API support
+  # sklearnex functional Array API support doesn't guaranty namespace consistency for the estimator's array attributes.
+  - decomposition/tests/test_pca.py::test_pca_array_api_compliance[PCA(n_components=2,svd_solver='covariance_eigh')-check_array_api_input_and_values-array_api_strict-None-None]
+  - decomposition/tests/test_pca.py::test_pca_array_api_compliance[PCA(n_components=2,svd_solver='covariance_eigh',whiten=True)-check_array_api_input_and_values-array_api_strict-None-None]
+  - decomposition/tests/test_pca.py::test_pca_array_api_compliance[PCA(n_components=2,svd_solver='covariance_eigh')-check_array_api_get_precision-array_api_strict-None-None]
+  - decomposition/tests/test_pca.py::test_pca_array_api_compliance[PCA(n_components=2,svd_solver='covariance_eigh',whiten=True)-check_array_api_get_precision-array_api_strict-None-None]
+  - linear_model/tests/test_ridge.py::test_ridge_array_api_compliance[Ridge(solver='svd')-check_array_api_attributes-array_api_strict-None-None]
+  - linear_model/tests/test_ridge.py::test_ridge_array_api_compliance[Ridge(solver='svd')-check_array_api_input_and_values-array_api_strict-None-None]
+  # `train_test_split` inconsistency for Array API inputs.
+  - model_selection/tests/test_split.py::test_array_api_train_test_split[True-None-array_api_strict-None-None]
+  - model_selection/tests/test_split.py::test_array_api_train_test_split[True-stratify1-array_api_strict-None-None]
+  - model_selection/tests/test_split.py::test_array_api_train_test_split[False-None-array_api_strict-None-None]
+  # PCA. Array API functionally supported for all factorizations. power_iteration_normalizer=["LU", "QR"]
+  - decomposition/tests/test_pca.py::test_array_api_error_and_warnings_on_unsupported_params
+  # PCA. InvalidParameterError: The 'M' parameter of randomized_svd must be an instance of 'numpy.ndarray' or a sparse matrix.
+  - decomposition/tests/test_pca.py::test_pca_array_api_compliance[PCA(n_components=2,power_iteration_normalizer='QR',random_state=0,svd_solver='randomized')-check_array_api_input_and_values-array_api_strict-None-None]
+  - decomposition/tests/test_pca.py::test_pca_array_api_compliance[PCA(n_components=2,power_iteration_normalizer='QR',random_state=0,svd_solver='randomized')-check_array_api_get_precision-array_api_strict-None-None]
+  # Ridge regression. Array API functionally supported for all solvers. Not raising error for non-svd solvers.
+  - linear_model/tests/test_ridge.py::test_array_api_error_and_warnings_for_solver_parameter[array_api_strict]
+
   # 'kulsinski' distance was deprecated in scipy 1.11 but still marked as supported in scikit-learn < 1.3
   - neighbors/tests/test_neighbors.py::test_kneighbors_brute_backend[float64-kulsinski] <1.3
   - neighbors/tests/test_neighbors.py::test_radius_neighbors_brute_backend[kulsinski] <1.3

requirements-test.txt

@@ -11,4 +11,5 @@ xgboost==2.1.1
 lightgbm==4.5.0
 catboost==1.2.7 ; python_version < '3.11' # TODO: Remove 3.11 condition when catboost supports numpy 2.0
 shap==0.46.0
+array-api-compat==1.8.0
 array-api-strict==2.0.1