scikit-learn · glemaitre · Dec 7, 2023 · Nov 5, 2022 · Nov 5, 2022 · Nov 6, 2022
diff --git a/doc/whats_new/v1.4.rst b/doc/whats_new/v1.4.rst
@@ -424,6 +424,17 @@ Changelog
   support missing values if all `estimators` support missing values.
   :pr:`27710` by :user:`Guillaume Lemaitre <glemaitre>`.
 
+:mod:`sklearn.feature_extraction`
+.................................
+
+- |API| Changed error type from :class:`AttributeError` to
+  :class:`exceptions.NotFittedError` in unfitted instances of
+  :class:`feature_extraction.DictVectorizer` for the following methods:
+  :func:`feature_extraction.DictVectorizer.inverse_transform`,
+  :func:`feature_extraction.DictVectorizer.restrict`,
+  :func:`feature_extraction.DictVectorizer.transform`.
+  :pr:`24838` by :user:`Lorenz Hertel <LoHertel>`.
+
 :mod:`sklearn.feature_selection`
 ................................
 

diff --git a/sklearn/feature_extraction/_dict_vectorizer.py b/sklearn/feature_extraction/_dict_vectorizer.py
@@ -338,6 +338,8 @@ def inverse_transform(self, X, dict_type=dict):
         D : list of dict_type objects of shape (n_samples,)
             Feature mappings for the samples in X.
         """
+        check_is_fitted(self, "feature_names_")
+
         # COO matrix is not subscriptable
         X = check_array(X, accept_sparse=["csr", "csc"])
         n_samples = X.shape[0]
@@ -373,6 +375,7 @@ def transform(self, X):
         Xa : {array, sparse matrix}
             Feature vectors; always 2-d.
         """
+        check_is_fitted(self, ["feature_names_", "vocabulary_"])
         return self._transform(X, fitting=False)
 
     def get_feature_names_out(self, input_features=None):
@@ -428,6 +431,8 @@ def restrict(self, support, indices=False):
         >>> v.get_feature_names_out()
         array(['bar', 'foo'], ...)
         """
+        check_is_fitted(self, "feature_names_")
+
         if not indices:
             support = np.where(support)[0]
 

diff --git a/sklearn/feature_extraction/tests/test_dict_vectorizer.py b/sklearn/feature_extraction/tests/test_dict_vectorizer.py
@@ -9,6 +9,7 @@
 import scipy.sparse as sp
 from numpy.testing import assert_allclose, assert_array_equal
 
+from sklearn.exceptions import NotFittedError
 from sklearn.feature_extraction import DictVectorizer
 from sklearn.feature_selection import SelectKBest, chi2
 
@@ -239,3 +240,23 @@ def test_dict_vectorizer_get_feature_names_out():
     assert isinstance(feature_names, np.ndarray)
     assert feature_names.dtype == object
     assert_array_equal(feature_names, ["1", "2", "3"])
+
+
+@pytest.mark.parametrize(
+    "method, input",
+    [
+        ("transform", [{1: 2, 3: 4}, {2: 4}]),
+        ("inverse_transform", [{1: 2, 3: 4}, {2: 4}]),
+        ("restrict", [True, False, True]),
+    ],
+)
+def test_dict_vectorizer_not_fitted_error(method, input):
+    """Check that unfitted DictVectorizer instance raises NotFittedError.
+
+    This should be part of the common test but currently they test estimator accepting
+    text input.
+    """
+    dv = DictVectorizer(sparse=False)
+
+    with pytest.raises(NotFittedError):
+        getattr(dv, method)(input)