From a9c6ad9baf878015653569109091828ceaf2db8e Mon Sep 17 00:00:00 2001
From: Adrin Jalali <adrin.jalali@gmail.com>
Date: Wed, 5 Sep 2018 17:55:07 +0200
Subject: [PATCH] [MRG+1] break the tie in Meanshift in case cluster
 intensities are the same (#11901)

---
 doc/whats_new/v0.20.rst                  |  6 ++++++
 sklearn/cluster/mean_shift_.py           |  8 +++++---
 sklearn/cluster/tests/test_mean_shift.py | 12 ++++++++++++
 3 files changed, 23 insertions(+), 3 deletions(-)

diff --git a/doc/whats_new/v0.20.rst b/doc/whats_new/v0.20.rst
index 2ed336b782174..46b262896145c 100644
--- a/doc/whats_new/v0.20.rst
+++ b/doc/whats_new/v0.20.rst
@@ -63,6 +63,7 @@ parameters, may produce different models from the previous version. This often
 occurs due to changes in the modelling logic (bug fixes or enhancements), or in
 random sampling procedures.
 
+- :class:`cluster.MeanShift` (bug fix)
 - :class:`decomposition.IncrementalPCA` in Python 2 (bug fix)
 - :class:`decomposition.SparsePCA` (bug fix)
 - :class:`ensemble.GradientBoostingClassifier` (bug fix affecting feature importances)
@@ -151,6 +152,11 @@ Support for Python 3.3 has been officially dropped.
   ``n_iter_`` attribute in the docstring of :class:`cluster.KMeans`.
   :issue:`11353` by :user:`Jeremie du Boisberranger <jeremiedbb>`.
 
+- |Fix| Fixed a bug in :func:`cluster.mean_shift` where the assigned labels
+  were not deterministic if there were multiple clusters with the same
+  intensities.
+  :issue:`11901` by :user:`Adrin Jalali <adrinjalali>`.
+
 - |API| Deprecate ``pooling_func`` unused parameter in
   :class:`cluster.AgglomerativeClustering`.
   :issue:`9875` by :user:`Kumar Ashutosh <thechargedneutron>`.
diff --git a/sklearn/cluster/mean_shift_.py b/sklearn/cluster/mean_shift_.py
index 487545ac039d3..800c85c365988 100644
--- a/sklearn/cluster/mean_shift_.py
+++ b/sklearn/cluster/mean_shift_.py
@@ -215,8 +215,10 @@ def mean_shift(X, bandwidth=None, seeds=None, bin_seeding=False,
     # If the distance between two kernels is less than the bandwidth,
     # then we have to remove one because it is a duplicate. Remove the
     # one with fewer points.
+
     sorted_by_intensity = sorted(center_intensity_dict.items(),
-                                 key=lambda tup: tup[1], reverse=True)
+                                 key=lambda tup: (tup[1], tup[0]),
+                                 reverse=True)
     sorted_centers = np.array([tup[0] for tup in sorted_by_intensity])
     unique = np.ones(len(sorted_centers), dtype=np.bool)
     nbrs = NearestNeighbors(radius=bandwidth,
@@ -359,9 +361,9 @@ class MeanShift(BaseEstimator, ClusterMixin):
     ...               [4, 7], [3, 5], [3, 6]])
     >>> clustering = MeanShift(bandwidth=2).fit(X)
     >>> clustering.labels_
-    array([0, 0, 0, 1, 1, 1])
+    array([1, 1, 1, 0, 0, 0])
     >>> clustering.predict([[0, 0], [5, 5]])
-    array([0, 1])
+    array([1, 0])
     >>> clustering # doctest: +NORMALIZE_WHITESPACE
     MeanShift(bandwidth=2, bin_seeding=False, cluster_all=True, min_bin_freq=1,
          n_jobs=None, seeds=None)
diff --git a/sklearn/cluster/tests/test_mean_shift.py b/sklearn/cluster/tests/test_mean_shift.py
index 1d6940a947dc2..441f822cdbded 100644
--- a/sklearn/cluster/tests/test_mean_shift.py
+++ b/sklearn/cluster/tests/test_mean_shift.py
@@ -101,6 +101,18 @@ def test_unfitted():
     assert_false(hasattr(ms, "labels_"))
 
 
+def test_cluster_intensity_tie():
+    X = np.array([[1, 1], [2, 1], [1, 0],
+                  [4, 7], [3, 5], [3, 6]])
+    c1 = MeanShift(bandwidth=2).fit(X)
+
+    X = np.array([[4, 7], [3, 5], [3, 6],
+                  [1, 1], [2, 1], [1, 0]])
+    c2 = MeanShift(bandwidth=2).fit(X)
+    assert_array_equal(c1.labels_, [1, 1, 1, 0, 0, 0])
+    assert_array_equal(c2.labels_, [0, 0, 0, 1, 1, 1])
+
+
 def test_bin_seeds():
     # Test the bin seeding technique which can be used in the mean shift
     # algorithm