scikit-learn
diff --git a/‎sklearn/neighbors/_kde.py
Copy file name to clipboardExpand all lines: sklearn/neighbors/_kde.py
+5-5Lines changed: 5 additions & 5 deletions b/‎sklearn/neighbors/_kde.py
Copy file name to clipboardExpand all lines: sklearn/neighbors/_kde.py
+5-5Lines changed: 5 additions & 5 deletions
diff --git a/‎sklearn/neighbors/tests/test_kde.py
Copy file name to clipboardExpand all lines: sklearn/neighbors/tests/test_kde.py
+15Lines changed: 15 additions & 0 deletions b/‎sklearn/neighbors/tests/test_kde.py
Copy file name to clipboardExpand all lines: sklearn/neighbors/tests/test_kde.py
+15Lines changed: 15 additions & 0 deletions
@@ -202,11 +202,6 @@ def _choose_algorithm(self, algorithm, metric):
             return algorithm
 
     def _evaluate_hypercube_faces(self):
-        lower, upper = np.transpose(self.bounds)
-        if np.any(upper <= lower):
-            raise ValueError(
-                f"invalid bounds: upper ({upper}) is not larger than lower ({lower})"
-            )
         n_dims = len(self.bounds)
         # Start the queue with all vertices, then recursively create all faces.
         queue = set(itertools.product(*self.bounds))
@@ -434,6 +429,11 @@ def _validate_data(
     ):
         if self.bounds is not None:
             lower, upper = np.transpose(self.bounds)
+            if np.any(upper <= lower):
+                raise ValueError(
+                    f"invalid bounds: upper ({upper}) is not larger than lower "
+                    f"({lower})"
+                )
             if np.any(X < lower):
                 raise ValueError("samples must be larger than lower bound")
             if np.any(X > upper):
 
@@ -274,6 +274,21 @@ def test_kde_hypercube_faces(n_dims, expected_counts):
     np.testing.assert_array_equal(actual_counts, expected_counts)
 
 
+def test_kde_with_bounds_invalid():
+    with pytest.raises(ValueError, match="invalid bounds"):
+        KernelDensity(bounds=[(3, 1)]).fit(None)
+
+    kde = KernelDensity(bounds=[(0, 2)]).fit(np.ones((3, 1)))
+    with pytest.raises(NotImplementedError):
+        kde.sample()
+
+    with pytest.raises(ValueError, match="samples must be larger"):
+        KernelDensity(bounds=[(1, 2)]).fit(np.zeros((3, 1)))
+
+    with pytest.raises(ValueError, match="samples must be smaller"):
+        KernelDensity(bounds=[(1, 2)]).fit(3 * np.ones((3, 1)))
+
+
 def test_kde_bounded_density():
     n_samples, n_features = (100_000, 2)
     rng = np.random.RandomState(0)