Apply new input validation to cuml.kernel_ridge

python/cuml/cuml/kernel_ridge/kernel_ridge.py

@@ -9,7 +9,6 @@
 from cupy import linalg
 from cupyx import geterr, lapack, seterr
 
-from cuml.common import input_to_cuml_array
 from cuml.common.array_descriptor import CumlArrayDescriptor
 from cuml.common.doc_utils import generate_docstring
 from cuml.internals import reflect
@@ -22,7 +21,7 @@
     to_gpu,
 )
 from cuml.internals.mixins import RegressorMixin
-from cuml.internals.validation import check_features, check_is_fitted
+from cuml.internals.validation import check_inputs, check_is_fitted
 from cuml.metrics import pairwise_kernels
 
 
@@ -53,7 +52,7 @@ def _solve_cholesky_kernel(K, y, alpha, sample_weight=None):
     n_samples = K.shape[0]
     n_targets = y.shape[1]
 
-    K = cp.array(K, dtype=np.float64)
+    K = cp.asarray(K, dtype=np.float64)
 
     alpha = cp.atleast_1d(alpha)
     one_alpha = alpha.size == 1
@@ -266,6 +265,10 @@ def __init__(
         self.coef0 = coef0
         self.kernel_params = kernel_params
 
+    @staticmethod
+    def _more_static_tags():
+        return {"multioutput": True}
+
     def _get_kernel(self, X, Y=None):
         if isinstance(self.kernel, str):
             params = {
@@ -277,42 +280,35 @@ def _get_kernel(self, X, Y=None):
             params = self.kernel_params or {}
         return pairwise_kernels(
             X, Y, metric=self.kernel, filter_params=True, **params
-        )
+        ).to_output("cupy")
 
     @generate_docstring()
-    @reflect(reset=True)
+    @reflect(reset="type")
     def fit(
         self, X, y, sample_weight=None, *, convert_dtype=True
     ) -> "KernelRidge":
-        ravel = False
-        if len(y.shape) == 1:
-            y = y.reshape(-1, 1)
-            ravel = True
-
-        X_m = input_to_cuml_array(
+        X, y, sample_weight, index = check_inputs(
+            self,
             X,
-            convert_to_dtype=(np.float32 if convert_dtype else None),
-            check_dtype=[np.float32, np.float64],
-        ).array
-
-        y_m = input_to_cuml_array(
             y,
-            check_dtype=X_m.dtype,
-            convert_to_dtype=(X_m.dtype if convert_dtype else None),
-            check_rows=X_m.shape[0],
-        ).array
-
-        if X.shape[1] < 1:
-            raise ValueError("X matrix must have at least a column")
+            sample_weight,
+            dtype=("float32", "float64"),
+            convert_dtype=convert_dtype,
+            accept_multi_output=True,
+            return_index=True,
+            reset=True,
+        )
+        if ravel := (y.ndim == 1):
+            y = y.reshape(-1, 1)
 
-        K = self._get_kernel(X_m)
+        K = self._get_kernel(X)
         dual_coef = _solve_cholesky_kernel(
-            K, cp.asarray(y_m), cp.asarray(self.alpha), sample_weight
-        ).astype(X_m.dtype, copy=False)
+            K, y, cp.asarray(self.alpha), sample_weight
+        ).astype(X.dtype, copy=False)
         if ravel:
             dual_coef = dual_coef.ravel()
 
-        self.X_fit_ = X_m
+        self.X_fit_ = CumlArray(data=X, index=index)
         self.dual_coef_ = CumlArray(data=dual_coef)
         return self
 
@@ -335,17 +331,14 @@ def predict(self, X, *, convert_dtype=True):
             Returns predicted values.
         """
         check_is_fitted(self)
-        check_features(self, X)
-
-        dtype = self.X_fit_.dtype
-
-        X_m = input_to_cuml_array(
+        X = check_inputs(
+            self,
             X,
-            check_dtype=dtype,
-            convert_to_dtype=(dtype if convert_dtype else None),
-            check_cols=self.n_features_in_,
-        ).array
-
-        K = cp.asarray(self._get_kernel(X_m, self.X_fit_), dtype=dtype)
+            dtype=self.X_fit_.dtype,
+            convert_dtype=convert_dtype,
+        )
+        K = self._get_kernel(X, self.X_fit_.to_output("cupy")).astype(
+            X.dtype, copy=False
+        )
         dual_coef = self.dual_coef_.to_output("cupy")
-        return CumlArray(cp.dot(K, dual_coef))
+        return CumlArray(data=cp.dot(K, dual_coef))

python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml

@@ -806,18 +806,7 @@
 - reason: KernelRidge input handling and validation
   marker: cuml_accel_kernel_ridge_estimator_checks
   tests:
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_dtype_object]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_estimators_empty_data_messages]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_estimators_nan_inf]"
   - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_estimators_pickle(readonly_memmap=True)]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_regressor_data_not_an_array]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_regressors_train(readonly_memmap=True)]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_regressors_train(readonly_memmap=True,X_dtype=float32)]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_regressors_train]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_requires_y_none]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_sample_weights_not_an_array]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_sample_weights_pandas_series]"
-  - "sklearn.tests.test_common::test_estimators[KernelRidge()-check_supervised_y_no_nan]"
 - reason: cuml.accel does not support callable initialization for KMeans
   marker: cuml_accel_kmeans_callable_init
   tests:

python/cuml/tests/test_sklearn_compatibility.py

@@ -97,19 +97,7 @@
     },
     KernelRidge: {
         "check_estimator_tags_renamed": "No support for modern tags infrastructure",
-        "check_sample_weights_pandas_series": "KernelRidge does not handle pandas Series sample weights",
-        "check_sample_weights_not_an_array": "KernelRidge does not handle non-array sample weights",
-        "check_all_zero_sample_weights_error": "KernelRidge does not validate all-zero sample weights",
-        "check_dtype_object": "KernelRidge does not handle object dtype",
-        "check_estimators_empty_data_messages": "KernelRidge does not handle empty data",
-        "check_estimators_nan_inf": "KernelRidge does not check for NaN and inf",
-        "check_regressors_train": "KernelRidge does not handle list inputs",
-        "check_regressors_train(readonly_memmap=True)": "KernelRidge does not handle readonly memmap",
-        "check_regressors_train(readonly_memmap=True,X_dtype=float32)": "KernelRidge does not handle readonly memmap with float32",
         "check_regressor_data_not_an_array": "KernelRidge does not handle non-array data",
-        "check_supervised_y_2d": "KernelRidge does not handle 2D y",
-        "check_supervised_y_no_nan": "KernelRidge does not check for NaN in y",
-        "check_requires_y_none": "KernelRidge does not handle y=None",
     },
     LogisticRegression: {
         "check_estimator_tags_renamed": "No support for modern tags infrastructure",