Merge pull request #306 from IFCA-Advanced-Computing/feature-bws-test

Add Baumgartner-Weiss-Schindler test data drift method

Author: Jaime Céspedes Sisniega

README.md

@@ -335,8 +335,8 @@ The currently implemented detectors are listed in the following table.
     <td style="text-align: center; border: 1px solid grey; padding: 8px;"><a href="https://doi.org/10.1007/978-3-540-75488-6_27">Nishida and Yamauchi (2007)</a></td>
   </tr>
   <tr>
-    <td rowspan="17" style="text-align: center; border: 1px solid grey; padding: 8px;">Data drift</td>
-    <td rowspan="15" style="text-align: center; border: 1px solid grey; padding: 8px;">Batch</td>
+    <td rowspan="18" style="text-align: center; border: 1px solid grey; padding: 8px;">Data drift</td>
+    <td rowspan="16" style="text-align: center; border: 1px solid grey; padding: 8px;">Batch</td>
     <td rowspan="10" style="text-align: center; border: 1px solid grey; padding: 8px;">Distance based</td>
     <td style="text-align: center; border: 1px solid grey; padding: 8px;">U</td>
     <td style="text-align: center; border: 1px solid grey; padding: 8px;">N</td>
@@ -398,12 +398,18 @@ The currently implemented detectors are listed in the following table.
     <td style="text-align: center; border: 1px solid grey; padding: 8px;"><a href="https://doi.org/10.1057/jors.2008.144">Wu and Olson (2010)</a></td>
   </tr>
   <tr>
-    <td rowspan="5" style="text-align: center; border: 1px solid grey; padding: 8px;">Statistical test</td>
+    <td rowspan="6" style="text-align: center; border: 1px solid grey; padding: 8px;">Statistical test</td>
     <td style="text-align: center; border: 1px solid grey; padding: 8px;">U</td>
     <td style="text-align: center; border: 1px solid grey; padding: 8px;">C</td>
     <td style="text-align: center; border: 1px solid grey; padding: 8px;">Chi-square test</td>
     <td style="text-align: center; border: 1px solid grey; padding: 8px;"><a href="https://doi.org/10.1080/14786440009463897">Pearson (1900)</a></td>
   </tr>
+  <tr>
+    <td style="text-align: center; border: 1px solid grey; padding: 8px;">U</td>
+    <td style="text-align: center; border: 1px solid grey; padding: 8px;">N</td>
+    <td style="text-align: center; border: 1px solid grey; padding: 8px;">Baumgartner-Weiss-Schindler test</td>
+    <td style="text-align: center; border: 1px solid grey; padding: 8px;"><a href="https://doi.org/10.2307/2533862">Baumgartner et al. (1998)</a></td>
+  </tr>
   <tr>
     <td style="text-align: center; border: 1px solid grey; padding: 8px;">U</td>
     <td style="text-align: center; border: 1px solid grey; padding: 8px;">N</td>

docs/source/api_reference/detectors/data_drift/batch.md

@@ -49,6 +49,7 @@ The {mod}`frouros.detectors.data_drift.batch` module contains batch data drift d
     :template: class.md
 
     AndersonDarlingTest
+    BWSTest
     ChiSquareTest
     CVMTest
     KSTest

frouros/detectors/data_drift/__init__.py

@@ -3,6 +3,7 @@
 from .batch import (  # noqa: F401
     AndersonDarlingTest,
     BhattacharyyaDistance,
+    BWSTest,
     ChiSquareTest,
     CVMTest,
     EMD,

frouros/detectors/data_drift/batch/__init__.py

@@ -13,6 +13,7 @@
 )
 from .statistical_test import (
     AndersonDarlingTest,
+    BWSTest,
     ChiSquareTest,
     CVMTest,
     KSTest,
@@ -22,6 +23,7 @@
 
 __all__ = [
     "AndersonDarlingTest",
+    "BWSTest",
     "BhattacharyyaDistance",
     "ChiSquareTest",
     "CVMTest",

frouros/detectors/data_drift/batch/statistical_test/__init__.py

@@ -1,6 +1,7 @@
 """Data drift batch statistical test detection methods' init."""
 
 from .anderson_darling import AndersonDarlingTest
+from .bws import BWSTest
 from .chisquare import ChiSquareTest
 from .cvm import CVMTest
 from .ks import KSTest
@@ -9,6 +10,7 @@
 
 __all__ = [
     "AndersonDarlingTest",
+    "BWSTest",
     "ChiSquareTest",
     "CVMTest",
     "KSTest",

frouros/detectors/data_drift/batch/statistical_test/bws.py

@@ -0,0 +1,70 @@
+"""BWSTest (Baumgartner-Weiss-Schindler test) module."""
+
+from typing import Optional, Union
+
+import numpy as np  # type: ignore
+from scipy.stats import bws_test  # type: ignore
+
+from frouros.callbacks.batch.base import BaseCallbackBatch
+from frouros.detectors.data_drift.base import NumericalData, UnivariateData
+from frouros.detectors.data_drift.batch.statistical_test.base import (
+    BaseStatisticalTest,
+    StatisticalResult,
+)
+
+
+class BWSTest(BaseStatisticalTest):
+    """BWSTest (Baumgartner-Weiss-Schindler test) [baumgartner1998nonparametric]_ detector.
+
+    :param callbacks: callbacks, defaults to None
+    :type callbacks: Optional[Union[BaseCallbackBatch, list[BaseCallbackBatch]]]
+
+    :Note:
+    - Passing additional arguments to `scipy.stats.bws_test <https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.bws_test.html>`__ can be done using :func:`compare` kwargs.
+
+    :References:
+
+    .. [baumgartner1998nonparametric] Baumgartner, W., P. Weiß, and H. Schindler.
+        "A nonparametric test for the general two-sample problem."
+        Biometrics (1998): 1129-1135.
+
+    :Example:
+
+    >>> from frouros.detectors.data_drift import BWSTest
+    >>> import numpy as np
+    >>> np.random.seed(seed=31)
+    >>> X = np.random.normal(loc=0, scale=1, size=100)
+    >>> Y = np.random.normal(loc=1, scale=1, size=100)
+    >>> detector = BWSTest()
+    >>> _ = detector.fit(X=X)
+    >>> detector.compare(X=Y)[0]
+    StatisticalResult(statistic=29.942072035675395, p_value=0.0001)
+    """  # noqa: E501  # pylint: disable=line-too-long
+
+    def __init__(  # noqa: D107
+        self,
+        callbacks: Optional[Union[BaseCallbackBatch, list[BaseCallbackBatch]]] = None,
+    ) -> None:
+        super().__init__(
+            data_type=NumericalData(),
+            statistical_type=UnivariateData(),
+            callbacks=callbacks,
+        )
+
+    @staticmethod
+    def _statistical_test(
+        X_ref: np.ndarray,  # noqa: N803
+        X: np.ndarray,
+        **kwargs,
+    ) -> StatisticalResult:
+        test = bws_test(
+            x=X_ref,
+            y=X,
+            alternative=kwargs.get("alternative", "two-sided"),
+            method=kwargs.get("method", None),
+        )
+        test = StatisticalResult(
+            statistic=test.statistic,
+            p_value=test.pvalue,
+        )
+        return test

frouros/tests/integration/test_callback.py

@@ -27,6 +27,7 @@
 from frouros.detectors.concept_drift.base import BaseConceptDrift
 from frouros.detectors.data_drift.batch import (
     AndersonDarlingTest,
+    BWSTest,
     BhattacharyyaDistance,
     CVMTest,
     EMD,
@@ -137,7 +138,14 @@ def test_batch_permutation_test_conservative(
 
 @pytest.mark.parametrize(
     "detector_class",
-    [AndersonDarlingTest, CVMTest, KSTest, MannWhitneyUTest, WelchTTest],
+    [
+        AndersonDarlingTest,
+        BWSTest,
+        CVMTest,
+        KSTest,
+        MannWhitneyUTest,
+        WelchTTest,
+    ],
 )
 def test_batch_reset_on_statistical_test_data_drift(
     X_ref_univariate,  # noqa: N803
@@ -153,6 +161,8 @@ def test_batch_reset_on_statistical_test_data_drift(
     :type X_test_univariate: numpy.ndarray
     :param detector_class: detector distance
     :type detector_class: BaseDataDriftBatch
+    :param mocker: mocker
+    :type mocker: pytest_mock.mocker
     """
     mocker.patch("frouros.detectors.data_drift.batch.base.BaseDataDriftBatch.reset")
 

frouros/tests/integration/test_data_drift.py

@@ -4,7 +4,17 @@
 
 import numpy as np  # type: ignore
 import pytest  # type: ignore
+from scipy.stats import PermutationMethod  # type: ignore
 
+from frouros.detectors.data_drift.batch import (
+    AndersonDarlingTest,
+    BWSTest,
+    ChiSquareTest,
+    CVMTest,
+    KSTest,
+    MannWhitneyUTest,
+    WelchTTest,
+)
 from frouros.detectors.data_drift.batch import (
     BhattacharyyaDistance,
     EMD,
@@ -16,14 +26,6 @@
     KL,
     MMD,
 )
-from frouros.detectors.data_drift.batch import (
-    AndersonDarlingTest,
-    ChiSquareTest,
-    CVMTest,
-    KSTest,
-    MannWhitneyUTest,
-    WelchTTest,
-)
 from frouros.detectors.data_drift.batch.base import BaseDataDriftBatch
 from frouros.detectors.data_drift.streaming import (  # noqa: N811
     IncrementalKSTest,
@@ -161,20 +163,32 @@ def test_batch_distance_bins_based_univariate_same_distribution(
 
 
 @pytest.mark.parametrize(
-    "detector, expected_statistic, expected_p_value",
+    "detector, expected_statistic, expected_p_value, kwargs",
     [
-        (AndersonDarlingTest(), 23171.19994366, 0.001),
-        (CVMTest(), 3776.09848103, 5.38105056e-07),
-        (KSTest(), 0.99576271, 0.0),
-        (MannWhitneyUTest(), 6912.0, 0.0),
-        (WelchTTest(), -287.92032554, 0.0),
+        (AndersonDarlingTest(), 23171.19994366, 0.001, {}),
+        (
+            BWSTest(),
+            108757.63520694,
+            0.00990099,
+            {
+                "method": PermutationMethod(
+                    n_resamples=100,
+                    random_state=31,
+                ),
+            },
+        ),
+        (CVMTest(), 3776.09848103, 5.38105056e-07, {}),
+        (KSTest(), 0.99576271, 0.0, {}),
+        (MannWhitneyUTest(), 6912.0, 0.0, {}),
+        (WelchTTest(), -287.92032554, 0.0, {}),
     ],
 )
 def test_batch_statistical_univariate(
     elec2_dataset: Tuple[np.ndarray, np.ndarray, np.ndarray],
     detector: BaseDataDriftBatch,
     expected_statistic: float,
     expected_p_value: float,
+    kwargs: dict,
 ) -> None:
     """Test statistical univariate method.
 
@@ -186,11 +200,13 @@ def test_batch_statistical_univariate(
     :type expected_statistic: float
     :param expected_p_value: expected p-value value
     :type expected_p_value: float
+    :param kwargs: additional arguments
+    :type kwargs: dict
     """
     X_ref, _, X_test = elec2_dataset  # noqa: N806
 
     _ = detector.fit(X=X_ref[:, 0])
-    (statistic, p_value), _ = detector.compare(X=X_test[:, 0])
+    (statistic, p_value), _ = detector.compare(X=X_test[:, 0], **kwargs)
 
     assert np.isclose(statistic, expected_statistic)
     assert np.isclose(p_value, expected_p_value)

pyproject.toml

@@ -39,7 +39,7 @@ dependencies = [
     "matplotlib>=3.8.2,<3.9",
     "numpy>=1.26.3,<1.27",
     "requests>=2.31.0,<2.32",
-    "scipy>=1.11.4,<1.13",
+    "scipy>=1.12.0,<1.13",
     "tqdm>=4.66.1,<5.0",
 ]