Add OutlierRemover and tests

.appveyor.yml

@@ -17,7 +17,7 @@ install:
   - conda config --set always_yes yes --set changeps1 no
   - conda create -n sklego-env -c pytorch %REQUIREMENTS% python=%PYTHON_VERSION%
   - activate sklego-env
-  - pip install -e .
+  - pip install -e .[dev]
 
 test_script:
   - activate sklego-env

setup.py

@@ -9,7 +9,7 @@
 docs_packages = ["sphinx>=1.8.5", "sphinx_rtd_theme>=0.4.3"]
 dev_packages = docs_packages + ["flake8>=3.6.0", "matplotlib>=3.0.2", "pytest>=4.0.2",
                                 "nbval>=0.9.1", "plotnine>=0.5.1", "jupyter>=1.0.0",
-                                "jupyterlab>=0.35.4", "pytest-cov>=2.6.1"]
+                                "jupyterlab>=0.35.4", "pytest-cov>=2.6.1", "pytest-mock>=1.6.3"]
 
 
 def read(fname):

sklego/common.py

@@ -45,9 +45,12 @@ class TrainOnlyTransformerMixin(TransformerMixin):
 
     }
 
-    def fit(self, X, y):
+    def fit(self, X, y=None):
         """Calculates the hash of X_train"""
-        check_X_y(X, y, estimator=self)
+        if y is None:
+            check_array(X, estimator=self)
+        else:
+            check_X_y(X, y, estimator=self)
         self.X_hash_ = self._hash(X)
         self.dim_ = X.shape[1]
         return self

sklego/meta.py

@@ -4,7 +4,7 @@
 from sklearn.base import BaseEstimator, TransformerMixin, MetaEstimatorMixin
 from sklearn.utils.validation import check_is_fitted, check_X_y, check_array, FLOAT_DTYPES
 
-from sklego.common import as_list
+from sklego.common import as_list, TrainOnlyTransformerMixin
 
 
 class EstimatorTransformer(TransformerMixin, MetaEstimatorMixin, BaseEstimator):
@@ -119,6 +119,32 @@ def predict(self, X):
             raise ValueError(f"found a group(s) {culprits} in `.predict` that was not in `.fit`")
 
 
+class OutlierRemover(TrainOnlyTransformerMixin, BaseEstimator):
+    """
+    Removes outliers (train-time only) using the supplied removal model.
+
+    :param outlier_detector: must implement `fit` and `predict` methods
+    :param refit: If True, fits the estimator during pipeline.fit().
+
+    """
+    def __init__(self, outlier_detector, refit=True):
+        self.outlier_detector = outlier_detector
+        self.refit = refit
+
+    def fit(self, X, y=None):
+        self.estimator_ = clone(self.outlier_detector)
+        if self.refit:
+            super().fit(X, y)
+            self.estimator_.fit(X, y)
+        return self
+
+    def transform_train(self, X):
+        check_is_fitted(self, 'estimator_')
+        predictions = self.estimator_.predict(X)
+        check_array(predictions, estimator=self.outlier_detector)
+        return X[predictions.squeeze() != -1]
+
+
 class DecayEstimator(BaseEstimator):
     """
     Morphs an estimator suchs that the training weights can be

tests/test_meta/test_outlier_remover.py

@@ -0,0 +1,48 @@
+import pytest
+from pandas.tests.extension.numpy_.test_numpy_nested import np
+from sklearn.utils import estimator_checks
+
+from sklego.common import flatten
+from sklego.meta import OutlierRemover
+from sklego.mixture import GMMOutlierDetector
+
+
+@pytest.mark.parametrize("test_fn", flatten([
+    estimator_checks.check_transformers_unfitted,
+    estimator_checks.check_fit2d_predict1d,
+    estimator_checks.check_fit2d_1sample,
+    estimator_checks.check_fit2d_1feature,
+    estimator_checks.check_fit1d,
+    estimator_checks.check_get_params_invariance,
+    estimator_checks.check_set_params,
+    estimator_checks.check_dont_overwrite_parameters,
+    estimator_checks.check_transformers_unfitted
+]))
+def test_estimator_checks(test_fn):
+    outlier_remover = OutlierRemover(outlier_detector=GMMOutlierDetector(), refit=True)
+    test_fn(OutlierRemover.__name__, outlier_remover)
+
+
+@pytest.fixture
+def mock_outlier_detector(mocker):
+    return mocker.Mock()
+
+
+def test_no_outliers(mock_outlier_detector, mocker):
+    mock_outlier_detector.fit.return_value = None
+    mock_outlier_detector.predict.return_value = np.array([[1, 1]])
+    mocker.patch('sklego.meta.clone').return_value = mock_outlier_detector
+
+    outlier_remover = OutlierRemover(outlier_detector=mock_outlier_detector, refit=True)
+    outlier_remover.fit(X=np.array([[1, 1], [2, 2]]))
+    assert len(outlier_remover.transform_train(np.array([[1, 1], [2, 2]]))) == 2
+
+
+def test_remove_outlier(mock_outlier_detector, mocker):
+    mock_outlier_detector.fit.return_value = None
+    mock_outlier_detector.predict.return_value = np.array([[-1]])
+    mocker.patch('sklego.meta.clone').return_value = mock_outlier_detector
+
+    outlier_remover = OutlierRemover(outlier_detector=mock_outlier_detector, refit=True)
+    outlier_remover.fit(X=np.array([[5, 5]]))
+    assert len(outlier_remover.transform_train(np.array([[0, 0]]))) == 0