BUG: drop_duplicates drops non-duplicate rows in the presence of integer columns (GH 11376)

Author: evanpw

doc/source/whatsnew/v0.17.1.txt

@@ -92,7 +92,7 @@ Bug Fixes
 - Bug in ``pivot_table`` with ``margins=True`` when indexes are of ``Categorical`` dtype (:issue:`10993`)
 - Bug in ``DataFrame.plot`` cannot use hex strings colors (:issue:`10299`)
 
-
+- Bug in ``DataFrame.drop_duplicates`` (regression from 0.16.2) causing some non-duplicate rows containing integer values to be dropped (:issue:`11376`)
 
 
 - Bug in ``pd.eval`` where unary ops in a list error (:issue:`11235`)

pandas/tests/test_frame.py

@@ -8380,6 +8380,25 @@ def test_drop_duplicates(self):
         expected = df.iloc[[-2,-1]]
         assert_frame_equal(result, expected)
 
+        # GH 11376
+        df = pd.DataFrame({'x': [7, 6, 3, 3, 4, 8, 0],
+                           'y': [0, 6, 5, 5, 9, 1, 2]})
+        expected = df.loc[df.index != 3]
+        assert_frame_equal(df.drop_duplicates(), expected)
+
+        df = pd.DataFrame([[1 , 0], [0, 2]])
+        assert_frame_equal(df.drop_duplicates(), df)
+
+        df = pd.DataFrame([[-2, 0], [0, -4]])
+        assert_frame_equal(df.drop_duplicates(), df)
+
+        x = np.iinfo(np.int64).max / 3 * 2
+        df = pd.DataFrame([[-x, x], [0, x + 4]])
+        assert_frame_equal(df.drop_duplicates(), df)
+
+        df = pd.DataFrame([[-x, x], [x, x + 4]])
+        assert_frame_equal(df.drop_duplicates(), df)
+
     def test_drop_duplicates_for_take_all(self):
         df = DataFrame({'AAA': ['foo', 'bar', 'baz', 'bar',
                                 'foo', 'bar', 'qux', 'foo'],