BUG: _nsorted incorrect with duplicated values in index (#13412)

Add note to whatsnew

Add nlargest benchmark

Author: mroeschke

asv_bench/benchmarks/frame_methods.py

@@ -1012,3 +1012,13 @@ def setup(self):
 
     def time_frame_quantile_axis1(self):
         self.df.quantile([0.1, 0.5], axis=1)
+
+class frame_nlargest(object):
+    goal_time = 0.2
+
+    def setup(self):
+        self.df = DataFrame(np.random.randn(1000, 3),
+                            columns=list('ABC'))
+
+    def time_frame_nlargest(self):
+        self.df.nlargest(100, 'A')

doc/source/whatsnew/v0.19.2.txt

@@ -51,7 +51,7 @@ Bug Fixes
 - Compat with python 3.6 for Timestamp pickles (:issue:`14689`)
 
 
-
+- Bug in ``DataFrame.nlargest`` and ``DataFrame.nsmallest`` when the index had duplicate values (:issue:`13412`)
 
 
 

pandas/core/frame.py

@@ -3395,9 +3395,9 @@ def _nsorted(self, columns, n, method, keep):
             columns = [columns]
         columns = list(columns)
         ser = getattr(self[columns[0]], method)(n, keep=keep)
-        ascending = dict(nlargest=False, nsmallest=True)[method]
-        return self.loc[ser.index].sort_values(columns, ascending=ascending,
-                                               kind='mergesort')
+        if isinstance(ser, Series):
+            ser = ser.to_frame()
+        return ser.merge(self, on=columns[0], left_index=True)[self.columns]
 
     def nlargest(self, n, columns, keep='first'):
         """Get the rows of a DataFrame sorted by the `n` largest

pandas/tests/frame/test_analytics.py

@@ -1323,6 +1323,34 @@ def test_nsmallest_multiple_columns(self):
         expected = df.sort_values(['a', 'c']).head(5)
         tm.assert_frame_equal(result, expected)
 
+    def test_nsmallest_nlargest_duplicate_index(self):
+        df = pd.DataFrame({'a': [1, 2, 3, 4],
+                           'b': [4, 3, 2, 1],
+                           'c': [0, 1, 2, 3]},
+                          index=[0, 0, 1, 1])
+        result = df.nsmallest(4, 'a')
+        expected = df.sort_values('a').head(4)
+        tm.assert_frame_equal(result, expected)
+
+        result = df.nlargest(4, 'a')
+        expected = df.sort_values('a', ascending=False).head(4)
+        tm.assert_frame_equal(result, expected)
+
+        result = df.nsmallest(4, ['a', 'c'])
+        expected = df.sort_values(['a', 'c']).head(4)
+        tm.assert_frame_equal(result, expected)
+
+        result = df.nsmallest(4, ['c', 'a'])
+        expected = df.sort_values(['c', 'a']).head(4)
+        tm.assert_frame_equal(result, expected)
+
+        result = df.nlargest(4, ['a', 'c'])
+        expected = df.sort_values(['a', 'c'], ascending=False).head(4)
+        tm.assert_frame_equal(result, expected)
+
+        result = df.nlargest(4, ['c', 'a'])
+        expected = df.sort_values(['c', 'a'], ascending=False).head(4)
+        tm.assert_frame_equal(result, expected)
     # ----------------------------------------------------------------------
     # Isin