update the doctests to be compatible with numpy>=2 (scikit-learn#29613)

Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>

Author: EmilyXinyi

sklearn/cluster/_mean_shift.py

@@ -82,7 +82,7 @@ def estimate_bandwidth(X, *, quantile=0.3, n_samples=None, random_state=0, n_job
     >>> X = np.array([[1, 1], [2, 1], [1, 0],
     ...               [4, 7], [3, 5], [3, 6]])
     >>> estimate_bandwidth(X, quantile=0.5)
-    1.61...
+    np.float64(1.61...)
     """
     X = check_array(X)
 

sklearn/conftest.py

@@ -209,7 +209,10 @@ def pytest_collection_modifyitems(config, items):
         )
         skip_doctests = True
 
-    if np_base_version >= parse_version("2"):
+    if np_base_version < parse_version("2"):
+        # TODO: configure numpy to output scalar arrays as regular Python scalars
+        # once possible to improve readability of the tests docstrings.
+        # https://numpy.org/neps/nep-0051-scalar-representation.html#implementation
         reason = "Due to NEP 51 numpy scalar repr has changed in numpy 2"
         skip_doctests = True
 

sklearn/covariance/_shrunk_covariance.py

@@ -335,7 +335,7 @@ def ledoit_wolf_shrinkage(X, assume_centered=False, block_size=1000):
     >>> X = rng.multivariate_normal(mean=[0, 0], cov=real_cov, size=50)
     >>> shrinkage_coefficient = ledoit_wolf_shrinkage(X)
     >>> shrinkage_coefficient
-    0.23...
+    np.float64(0.23...)
     """
     X = check_array(X)
     # for only one feature, the result is the same whatever the shrinkage
@@ -452,7 +452,7 @@ def ledoit_wolf(X, *, assume_centered=False, block_size=1000):
     array([[0.44..., 0.16...],
            [0.16..., 0.80...]])
     >>> shrinkage
-    0.23...
+    np.float64(0.23...)
     """
     estimator = LedoitWolf(
         assume_centered=assume_centered,
@@ -672,7 +672,7 @@ def oas(X, *, assume_centered=False):
     array([[0.7533..., 0.2763...],
            [0.2763..., 0.3964...]])
     >>> shrinkage
-    0.0195...
+    np.float64(0.0195...)
     """
     estimator = OAS(
         assume_centered=assume_centered,
@@ -778,7 +778,7 @@ class OAS(EmpiricalCovariance):
     array([[ 1.7833..., -1.2431... ],
            [-1.2431...,  3.3889...]])
     >>> oas.shrinkage_
-    0.0195...
+    np.float64(0.0195...)
     """
 
     @_fit_context(prefer_skip_nested_validation=True)

sklearn/datasets/_base.py

@@ -571,7 +571,7 @@ def load_wine(*, return_X_y=False, as_frame=False):
     >>> data.target[[10, 80, 140]]
     array([0, 1, 2])
     >>> list(data.target_names)
-    ['class_0', 'class_1', 'class_2']
+    [np.str_('class_0'), np.str_('class_1'), np.str_('class_2')]
     """
 
     data, target, target_names, fdescr = load_csv_data(
@@ -705,7 +705,7 @@ def load_iris(*, return_X_y=False, as_frame=False):
     >>> data.target[[10, 25, 50]]
     array([0, 0, 1])
     >>> list(data.target_names)
-    ['setosa', 'versicolor', 'virginica']
+    [np.str_('setosa'), np.str_('versicolor'), np.str_('virginica')]
 
     See :ref:`sphx_glr_auto_examples_datasets_plot_iris_dataset.py` for a more
     detailed example of how to work with the iris dataset.
@@ -833,7 +833,7 @@ def load_breast_cancer(*, return_X_y=False, as_frame=False):
     >>> data.target[[10, 50, 85]]
     array([0, 1, 0])
     >>> list(data.target_names)
-    ['malignant', 'benign']
+    [np.str_('malignant'), np.str_('benign')]
     """
     data_file_name = "breast_cancer.csv"
     data, target, target_names, fdescr = load_csv_data(

sklearn/datasets/_samples_generator.py

@@ -204,7 +204,7 @@ def make_classification(
     >>> y.shape
     (100,)
     >>> list(y[:5])
-    [0, 0, 1, 1, 0]
+    [np.int64(0), np.int64(0), np.int64(1), np.int64(1), np.int64(0)]
     """
     generator = check_random_state(random_state)
 
@@ -556,7 +556,8 @@ def make_hastie_10_2(n_samples=12000, *, random_state=None):
     >>> y.shape
     (24000,)
     >>> list(y[:5])
-    [-1.0, 1.0, -1.0, 1.0, -1.0]
+    [np.float64(-1.0), np.float64(1.0), np.float64(-1.0), np.float64(1.0),
+    np.float64(-1.0)]
     """
     rs = check_random_state(random_state)
 
@@ -796,7 +797,7 @@ def make_circles(
     >>> y.shape
     (100,)
     >>> list(y[:5])
-    [1, 1, 1, 0, 0]
+    [np.int64(1), np.int64(1), np.int64(1), np.int64(0), np.int64(0)]
     """
     if isinstance(n_samples, numbers.Integral):
         n_samples_out = n_samples // 2
@@ -1168,7 +1169,7 @@ def make_friedman1(n_samples=100, n_features=10, *, noise=0.0, random_state=None
     >>> y.shape
     (100,)
     >>> list(y[:3])
-    [16.8..., 5.8..., 9.4...]
+    [np.float64(16.8...), np.float64(5.8...), np.float64(9.4...)]
     """
     generator = check_random_state(random_state)
 
@@ -1250,7 +1251,7 @@ def make_friedman2(n_samples=100, *, noise=0.0, random_state=None):
     >>> y.shape
     (100,)
     >>> list(y[:3])
-    [1229.4..., 27.0..., 65.6...]
+    [np.float64(1229.4...), np.float64(27.0...), np.float64(65.6...)]
     """
     generator = check_random_state(random_state)
 
@@ -1334,7 +1335,7 @@ def make_friedman3(n_samples=100, *, noise=0.0, random_state=None):
     >>> y.shape
     (100,)
     >>> list(y[:3])
-    [1.5..., 0.9..., 0.4...]
+    [np.float64(1.5...), np.float64(0.9...), np.float64(0.4...)]
     """
     generator = check_random_state(random_state)
 
@@ -2049,7 +2050,7 @@ def make_gaussian_quantiles(
     >>> y.shape
     (100,)
     >>> list(y[:5])
-    [2, 0, 1, 0, 2]
+    [np.int64(2), np.int64(0), np.int64(1), np.int64(0), np.int64(2)]
     """
     if n_samples < n_classes:
         raise ValueError("n_samples must be at least n_classes")

sklearn/decomposition/_dict_learning.py

@@ -847,15 +847,15 @@ def dict_learning_online(
     We can check the level of sparsity of `U`:
 
     >>> np.mean(U == 0)
-    0.53...
+    np.float64(0.53...)
 
     We can compare the average squared euclidean norm of the reconstruction
     error of the sparse coded signal relative to the squared euclidean norm of
     the original signal:
 
     >>> X_hat = U @ V
     >>> np.mean(np.sum((X_hat - X) ** 2, axis=1) / np.sum(X ** 2, axis=1))
-    0.05...
+    np.float64(0.05...)
     """
     # TODO(1.6): remove in 1.6
     if max_iter is None:
@@ -1049,15 +1049,15 @@ def dict_learning(
     We can check the level of sparsity of `U`:
 
     >>> np.mean(U == 0)
-    0.6...
+    np.float64(0.6...)
 
     We can compare the average squared euclidean norm of the reconstruction
     error of the sparse coded signal relative to the squared euclidean norm of
     the original signal:
 
     >>> X_hat = U @ V
     >>> np.mean(np.sum((X_hat - X) ** 2, axis=1) / np.sum(X ** 2, axis=1))
-    0.01...
+    np.float64(0.01...)
     """
     estimator = DictionaryLearning(
         n_components=n_components,
@@ -1551,15 +1551,15 @@ class DictionaryLearning(_BaseSparseCoding, BaseEstimator):
     We can check the level of sparsity of `X_transformed`:
 
     >>> np.mean(X_transformed == 0)
-    0.52...
+    np.float64(0.52...)
 
     We can compare the average squared euclidean norm of the reconstruction
     error of the sparse coded signal relative to the squared euclidean norm of
     the original signal:
 
     >>> X_hat = X_transformed @ dict_learner.components_
     >>> np.mean(np.sum((X_hat - X) ** 2, axis=1) / np.sum(X ** 2, axis=1))
-    0.05...
+    np.float64(0.05...)
     """
 
     _parameter_constraints: dict = {
@@ -1914,15 +1914,15 @@ class MiniBatchDictionaryLearning(_BaseSparseCoding, BaseEstimator):
     We can check the level of sparsity of `X_transformed`:
 
     >>> np.mean(X_transformed == 0) > 0.5
-    True
+    np.True_
 
     We can compare the average squared euclidean norm of the reconstruction
     error of the sparse coded signal relative to the squared euclidean norm of
     the original signal:
 
     >>> X_hat = X_transformed @ dict_learner.components_
     >>> np.mean(np.sum((X_hat - X) ** 2, axis=1) / np.sum(X ** 2, axis=1))
-    0.052...
+    np.float64(0.052...)
     """
 
     _parameter_constraints: dict = {

sklearn/decomposition/_sparse_pca.py

@@ -267,7 +267,7 @@ class SparsePCA(_BaseSparsePCA):
     (200, 5)
     >>> # most values in the components_ are zero (sparsity)
     >>> np.mean(transformer.components_ == 0)
-    0.9666...
+    np.float64(0.9666...)
     """
 
     _parameter_constraints: dict = {
@@ -473,7 +473,7 @@ class MiniBatchSparsePCA(_BaseSparsePCA):
     (200, 5)
     >>> # most values in the components_ are zero (sparsity)
     >>> np.mean(transformer.components_ == 0)
-    0.9...
+    np.float64(0.9...)
     """
 
     _parameter_constraints: dict = {

sklearn/feature_selection/_from_model.py

@@ -219,7 +219,7 @@ class SelectFromModel(MetaEstimatorMixin, SelectorMixin, BaseEstimator):
     >>> selector.estimator_.coef_
     array([[-0.3252...,  0.8345...,  0.4976...]])
     >>> selector.threshold_
-    0.55249...
+    np.float64(0.55249...)
     >>> selector.get_support()
     array([False,  True, False])
     >>> selector.transform(X)

sklearn/isotonic.py

@@ -65,10 +65,10 @@ def check_increasing(x, y):
     >>> from sklearn.isotonic import check_increasing
     >>> x, y = [1, 2, 3, 4, 5], [2, 4, 6, 8, 10]
     >>> check_increasing(x, y)
-    True
+    np.True_
     >>> y = [10, 8, 6, 4, 2]
     >>> check_increasing(x, y)
-    False
+    np.False_
     """
 
     # Calculate Spearman rho estimate and set return accordingly.

sklearn/linear_model/_base.py

@@ -544,7 +544,7 @@ class LinearRegression(MultiOutputMixin, RegressorMixin, LinearModel):
     >>> reg.coef_
     array([1., 2.])
     >>> reg.intercept_
-    3.0...
+    np.float64(3.0...)
     >>> reg.predict(np.array([[3, 5]]))
     array([16.])
     """