Merge pull request scikit-learn#4650 from lesteve/fix-4641

[MRG + 1] Fix DBSCAN fit with precomputed matrix in edge cases

Author: ogrisel

sklearn/cluster/dbscan_.py

@@ -115,7 +115,8 @@ def dbscan(X, eps=0.5, min_samples=5, metric='minkowski',
     # neighborhood of point i. While True, its useless information)
     if metric == 'precomputed':
         D = pairwise_distances(X, metric=metric)
-        neighborhoods = np.array([np.where(x <= eps)[0] for x in D])
+        neighborhoods = np.empty(X.shape[0], dtype=object)
+        neighborhoods[:] = [np.where(x <= eps)[0] for x in D]
     else:
         neighbors_model = NearestNeighbors(radius=eps, algorithm=algorithm,
                                            leaf_size=leaf_size,

sklearn/cluster/tests/test_dbscan.py

@@ -293,3 +293,15 @@ def test_dbscan_core_samples_toy():
                                       min_samples=4)
         assert_array_equal(core_samples, [])
         assert_array_equal(labels, -np.ones(n_samples))
+
+
+def test_dbscan_precomputed_metric_with_degenerate_input_arrays():
+    # see https://github.com/scikit-learn/scikit-learn/issues/4641 for
+    # more details
+    X = np.ones((10, 2))
+    labels = DBSCAN(eps=0.5, metric='precomputed').fit(X).labels_
+    assert_equal(len(set(labels)), 1)
+
+    X = np.zeros((10, 2))
+    labels = DBSCAN(eps=0.5, metric='precomputed').fit(X).labels_
+    assert_equal(len(set(labels)), 1)