scikit-learn
diff --git a/‎sklearn/utils/tests/test_testing.py
Lines changed: 49 additions & 4 deletions b/‎sklearn/utils/tests/test_testing.py
Lines changed: 49 additions & 4 deletions
@@ -1,9 +1,9 @@
 import warnings
 import unittest
 import sys
+import numpy as np
 
 from nose.tools import assert_raises
-
 from sklearn.utils.testing import (
     _assert_less,
     _assert_greater,
@@ -13,10 +13,15 @@
     assert_no_warnings,
     assert_equal,
     set_random_state,
-    assert_raise_message)
-
+    assert_raise_message,
+    assert_same_model,
+    assert_not_same_model)
 from sklearn.tree import DecisionTreeClassifier
 from sklearn.lda import LDA
+from sklearn.qda import QDA
+from sklearn.datasets import make_blobs
+from sklearn.svm import LinearSVC
+from sklearn.cluster import KMeans
 
 try:
     from nose.tools import assert_less
@@ -96,10 +101,50 @@ def _no_raise():
                   "test", _no_raise)
 
 
+def test_assert_same_not_same_model():
+    X1, y1 = make_blobs(n_samples=200, n_features=5, center_box=(-200, -150),
+                        centers=2, random_state=0)
+    X2, y2 = make_blobs(n_samples=100, n_features=5, center_box=(-1, 1),
+                        centers=3, random_state=1)
+    X3, y3 = make_blobs(n_samples=50, n_features=5, center_box=(-100, -50),
+                        centers=4, random_state=2)
+
+    for Estimator in (LinearSVC, KMeans):
+        assert_same_model(X3, Estimator(random_state=0).fit(X1, y1),
+                          Estimator(random_state=0).fit(X1, y1))
+
+        assert_raises(AssertionError, assert_not_same_model, X3,
+                      Estimator(random_state=0).fit(X1, y1),
+                      Estimator(random_state=0).fit(X1, y1))
+
+        assert_raises(AssertionError, assert_same_model, X3,
+                      Estimator().fit(X1, y1), Estimator().fit(X2, y2))
+
+        assert_not_same_model(X3, Estimator().fit(X1, y1),
+                              Estimator().fit(X2, y2))
+
+
+def test_qda_same_model():
+    # NRT to make sure the rotations_ attribute is correctly compared
+    X = np.array([[-1, -1], [-2, -1], [-3, -2], [1, 1], [2, 1], [3, 2]])
+    y = np.array([1, 1, 1, 2, 2, 2])
+    X2 = np.array([[-9, -9], [-20, -10], [-2, -2], [1, 1], [20, 10], [30, 20]])
+    y2 = np.array([1, 1, 2, 2, 3, 3])
+
+    clf1 = QDA().fit(X, y)
+    clf2 = QDA().fit(X, y)
+    assert_same_model(X2, clf1, clf2)
+
+    X2 = np.array([[-9, -9], [-20, -10], [-2, -2], [1, 1], [20, 10], [30, 20]])
+    y2 = np.array([1, 1, 2, 2, 3, 3])
+    clf3 = QDA().fit(X2, y2)
+    assert_not_same_model(X + X2, clf1, clf3)
+
+
 # This class is inspired from numpy 1.7 with an alteration to check
 # the reset warning filters after calls to assert_warns.
 # This assert_warns behavior is specific to scikit-learn because
-#`clean_warning_registry()` is called internally by assert_warns
+# `clean_warning_registry()` is called internally by assert_warns
 # and clears all previous filters.
 class TestWarns(unittest.TestCase):
     def test_warn(self):