FIX Gaussian KDE should return array, not scalar

Fixes scikit-learn#3269.

Author: larsmans

sklearn/neighbors/kde.py

@@ -200,7 +200,7 @@ def sample(self, n_samples=1, random_state=None):
         i = rng.randint(data.shape[0], size=n_samples)
 
         if self.kernel == 'gaussian':
-            return rng.normal(data[i], self.bandwidth)
+            return np.atleast_2d(rng.normal(data[i], self.bandwidth))
 
         elif self.kernel == 'tophat':
             # we first draw points from a d-dimensional normal distribution,

sklearn/neighbors/tests/test_kde.py

@@ -79,6 +79,11 @@ def test_kernel_density_sampling(n_samples=100, n_features=3):
         kde = KernelDensity(bandwidth, kernel=kernel).fit(X)
         assert_raises(NotImplementedError, kde.sample, 100)
 
+    # non-regression test: used to return a scalar
+    X = np.random.random((4, 1))
+    kde = KernelDensity(kernel="gaussian").fit(X)
+    assert_equal(kde.sample().shape, (1, 1))
+
 
 def test_kde_algorithm_metric_choice():
     """Smoke test for various metrics and algorithms"""