A few sklearn compat test fixes

python/cuml/cuml/_thirdparty/sklearn/preprocessing/_data.py

@@ -24,7 +24,6 @@
 # This code is under BSD 3 clause license.
 # Authors mentioned above do not endorse or promote this production.
 
-import numbers
 import warnings
 from itertools import chain, combinations
 from itertools import combinations_with_replacement as combinations_w_r
@@ -772,10 +771,10 @@ def partial_fit(self, X, y=None) -> "StandardScaler":
         # if n_samples_seen_ is an integer (i.e. no missing values), we need to
         # transform it to a NumPy array of shape (n_features,) required by
         # incr_mean_variance_axis and _incremental_variance_axis
-        if (hasattr(self, 'n_samples_seen_') and
-                isinstance(self.n_samples_seen_, numbers.Integral)):
-            self.n_samples_seen_ = np.repeat(
-                self.n_samples_seen_, X.shape[1]).astype(np.int64, copy=False)
+        if hasattr(self, 'n_samples_seen_') and np.isscalar(self.n_samples_seen_):
+            self.n_samples_seen_ = np.full(
+                X.shape[1], self.n_samples_seen_, dtype="int64"
+            )
 
         if sparse.issparse(X):
             if self.with_mean:

python/cuml/cuml/decomposition/pca.pyx

@@ -479,6 +479,8 @@ class PCA(InteropMixin,
             accept_large_sparse=True,
             dtype=("float32", "float64"),
             convert_dtype=convert_dtype,
+            ensure_min_samples=2,
+            ensure_min_features=2,
             order="F",
             reset=True,
         )

python/cuml/cuml/decomposition/tsvd.pyx

@@ -313,6 +313,8 @@ class TruncatedSVD(InteropMixin,
             dtype=("float32", "float64"),
             convert_dtype=convert_dtype,
             order="F",
+            ensure_min_samples=2,
+            ensure_min_features=2,
             return_index=True,
             reset=True,
         )

python/cuml/cuml/manifold/t_sne.pyx

@@ -595,6 +595,8 @@ class TSNE(InteropMixin,
             convert_dtype=convert_dtype,
             order="F",
             accept_sparse="csr",
+            ensure_min_samples=2,
+            ensure_min_features=2,
             reset=True,
             return_index=True,
         )

python/cuml/cuml/neighbors/kneighbors_regressor.pyx

@@ -208,6 +208,11 @@ class KNeighborsRegressor(RegressorMixin, FMajorInputTagMixin, NeighborsBase):
         )
         self.weights = weights
 
+    def __sklearn_tags__(self):
+        tags = super().__sklearn_tags__()
+        tags.target_tags.multi_output = True
+        return tags
+
     @generate_docstring()
     @mlfunc(set_input_type=True)
     def fit(self, X, y, *, convert_dtype="deprecated") -> "KNeighborsRegressor":

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

@@ -183,7 +183,6 @@
   - "sklearn.linear_model.tests.test_sparse_coordinate_descent::test_sparse_dense_equality[csc_matrix-True-6-24-True-ElasticNet]"
   - "sklearn.linear_model.tests.test_sparse_coordinate_descent::test_sparse_dense_equality[csc_matrix-True-6-24-True-Lasso]"
   - "sklearn.utils.tests.test_estimator_checks::test_check_estimator"
-  - "sklearn.utils.tests.test_estimator_checks::test_check_estimator_clones"
 - reason: Test should fail with cuml.accel
   marker: cuml_accel_bugs
   tests:
@@ -813,12 +812,8 @@
   marker: cuml_accel_test_estimators
   tests:
   - "sklearn.tests.test_common::test_estimators[KNeighborsRegressor()-check_supervised_y_no_nan]"
-  - "sklearn.tests.test_common::test_estimators[PCA()-check_fit2d_1feature]"
-  - "sklearn.tests.test_common::test_estimators[PCA()-check_fit2d_1sample]"
   - "sklearn.tests.test_common::test_estimators[RandomForestClassifier()-check_classifiers_multilabel_output_format_decision_function]"
   - "sklearn.tests.test_common::test_estimators[RandomForestRegressor()-check_regressor_data_not_an_array]"
-  - "sklearn.tests.test_common::test_estimators[TruncatedSVD()-check_fit2d_1feature]"
-  - "sklearn.tests.test_common::test_estimators[TruncatedSVD()-check_fit2d_1sample]"
 - reason: test_estimators checks fail
   marker: cuml_accel_test_estimators
   strict: false

python/cuml/tests/test_sklearn_compatibility.py

@@ -182,11 +182,11 @@ def _all_cuml_estimators():
 
 XFAILS = {
     KMeans: {
-        "check_sample_weight_equivalence_on_dense_data": "KMeans sample weight equivalence not implemented",
+        "check_sample_weight_equivalence_on_dense_data": "Sample weights not equal to repeating data",
     },
     LogisticRegression: {
-        "check_sample_weight_equivalence_on_dense_data": "LogisticRegression sample weight equivalence not implemented",
-        "check_sample_weight_equivalence_on_sparse_data": "LogisticRegression does not handle sparse data",
+        "check_sample_weight_equivalence_on_dense_data": "Sample weights not equal to repeating data",
+        "check_sample_weight_equivalence_on_sparse_data": "Sample weights not equal to repeating data",
         "check_class_weight_classifiers": "LogisticRegression does not handle class weights properly",
     },
     Ridge: {
@@ -197,70 +197,63 @@ def _all_cuml_estimators():
             "RandomForest uses quantile-binned splits, so sample weighting is "
             "not equivalent to duplicating rows"
         ),
-        "check_sample_weight_equivalence_on_sparse_data": (
-            "RandomForestClassifier does not handle sparse data"
-        ),
     },
     RandomForestRegressor: {
-        "check_regressor_data_not_an_array": "RandomForestRegressor does not handle non-array data",
+        "check_regressor_data_not_an_array": (
+            "cuml defaults to float32 for non-arrays (while sklearn defaults to "
+            "float64). Our float32 and float64 results differ _just enough_ that "
+            "this test fails on tolerances."
+        ),
         "check_sample_weight_equivalence_on_dense_data": (
             "RandomForest uses quantile-binned splits, so sample weighting is "
             "not equivalent to duplicating rows"
         ),
-        "check_sample_weight_equivalence_on_sparse_data": (
-            "RandomForestRegressor does not handle sparse data"
-        ),
     },
     KNeighborsRegressor: {
-        "check_supervised_y_2d": "KNeighborsRegressor does not handle 2D y",
+        "check_regressor_multioutput": (
+            "predict returns float32 output, but the test expects float64"
+        ),
     },
     LinearSVC: {
-        "check_sample_weight_equivalence_on_dense_data": "LinearSVC sample weight equivalence not implemented",
+        "check_sample_weight_equivalence_on_dense_data": "Sample weights not equal to repeating data",
     },
     LinearSVR: {
-        "check_sample_weight_equivalence_on_dense_data": "LinearSVR sample weight equivalence not implemented",
+        "check_sample_weight_equivalence_on_dense_data": "Sample weights not equal to repeating data",
     },
     SVC: {
-        "check_sample_weight_equivalence_on_dense_data": "SVC sample weight equivalence not implemented",
-        "check_sample_weight_equivalence_on_sparse_data": "SVC does not handle sparse data",
+        "check_sample_weight_equivalence_on_dense_data": "Sample weights not equal to repeating data",
+        "check_sample_weight_equivalence_on_sparse_data": "Sample weights not equal to repeating data",
     },
     SVR: {
-        "check_sample_weight_equivalence_on_dense_data": "SVR sample weight equivalence not implemented",
-        "check_sample_weight_equivalence_on_sparse_data": "SVR does not handle sparse data",
-    },
-    PCA: {
-        "check_fit2d_1sample": "PCA does not handle single sample",
-        "check_fit2d_1feature": "PCA does not handle single feature",
-    },
-    TruncatedSVD: {
-        "check_fit2d_1sample": "TruncatedSVD does not handle single sample",
-        "check_fit2d_1feature": "TruncatedSVD does not handle single feature",
+        "check_sample_weight_equivalence_on_dense_data": "Sample weights not equal to repeating data",
+        "check_sample_weight_equivalence_on_sparse_data": "Sample weights not equal to repeating data",
     },
     TSNE: {
         "check_dont_overwrite_parameters": "TSNE only supports n_components = 2",
         "check_pipeline_consistency": "TSNE results are not deterministic",
         "check_methods_sample_order_invariance": "TSNE results depend on sample order",
         "check_methods_subset_invariance": "TSNE results depend on data subset",
-        "check_fit2d_1sample": "TSNE does not handle single sample",
-        "check_fit2d_1feature": "TSNE does not handle single feature",
         "check_fit2d_predict1d": "TSNE only supports n_components = 2",
     },
     UMAP: {
-        "check_transformer_data_not_an_array": "UMAP does not handle non-array data",
+        "check_transformer_data_not_an_array": (
+            "cuml defaults to float32 for non-arrays (while sklearn defaults to "
+            "float64). Our float32 and float64 results differ _just enough_ that "
+            "this test fails on tolerances."
+        ),
         "check_methods_sample_order_invariance": "UMAP results depend on sample order",
         "check_transformer_general": "UMAP does not have consistent fit_transform and transform outputs",
         "check_methods_subset_invariance": "UMAP results depend on data subset",
         "check_transformer_preserve_dtypes": "UMAP returns float32 embeddings",
     },
     Lasso: {
-        "check_sample_weight_equivalence_on_sparse_data": "Lasso QN solver has issues with sample weights",
+        "check_sample_weight_equivalence_on_sparse_data": "Sample weights not equal to repeating data",
     },
     ElasticNet: {
-        "check_sample_weight_equivalence_on_sparse_data": "ElasticNet QN solver has issues with sample weights",
+        "check_sample_weight_equivalence_on_sparse_data": "Sample weights not equal to repeating data",
     },
     StandardScaler: {
         "check_no_attributes_set_in_init": "Vendored __init__ sets copy/with_mean/with_std as attributes",
-        "check_fit_score_takes_y": "AttributeError: 'int' object has no attribute 'repeat'",
         "check_do_not_raise_errors_in_init_or_set_params": "StandardScaler(**params) raises an exception",
     },
 }