BUG: Fix strange behaviour of Series.iloc on MultiIndex Series (#17148)…

… (#17291)

doc/source/whatsnew/v0.21.0.txt

@@ -353,6 +353,7 @@ Indexing
 - Fixes ``DataFrame.loc`` for setting with alignment and tz-aware ``DatetimeIndex`` (:issue:`16889`)
 - Avoids ``IndexError`` when passing an Index or Series to ``.iloc`` with older numpy (:issue:`17193`)
 - Allow unicode empty strings as placeholders in multilevel columns in Python 2 (:issue:`17099`)
+- Bug in ``.iloc`` when used with inplace addition or assignment and an int indexer on a ``MultiIndex`` causing the wrong indexes to be read from and written to (:issue:`17148`)
 
 I/O
 ^^^

pandas/core/indexing.py

@@ -146,7 +146,8 @@ def _get_setitem_indexer(self, key):
             return self._convert_tuple(key, is_setter=True)
 
         axis = self.obj._get_axis(0)
-        if isinstance(axis, MultiIndex):
+
+        if isinstance(axis, MultiIndex) and self.name != 'iloc':
             try:
                 return axis.get_loc(key)
             except Exception:

pandas/tests/indexing/test_iloc.py

@@ -269,6 +269,35 @@ def test_iloc_setitem(self):
         expected = Series([0, 1, 0], index=[4, 5, 6])
         tm.assert_series_equal(s, expected)
 
+    @pytest.mark.parametrize(
+        'data, indexes, values, expected_k', [
+            # test without indexer value in first level of MultiIndex
+            ([[2, 22, 5], [2, 33, 6]], [0, -1, 1], [2, 3, 1], [7, 10]),
+            # test like code sample 1 in the issue
+            ([[1, 22, 555], [1, 33, 666]], [0, -1, 1], [200, 300, 100],
+                [755, 1066]),
+            # test like code sample 2 in the issue
+            ([[1, 3, 7], [2, 4, 8]], [0, -1, 1], [10, 10, 1000], [17, 1018]),
+            # test like code sample 3 in the issue
+            ([[1, 11, 4], [2, 22, 5], [3, 33, 6]], [0, -1, 1], [4, 7, 10],
+                [8, 15, 13])
+        ])
+    def test_iloc_setitem_int_multiindex_series(
+            self, data, indexes, values, expected_k):
+        # GH17148
+        df = pd.DataFrame(
+            data=data,
+            columns=['i', 'j', 'k'])
+        df = df.set_index(['i', 'j'])
+
+        series = df.k.copy()
+        for i, v in zip(indexes, values):
+            series.iloc[i] += v
+
+        df['k'] = expected_k
+        expected = df.k
+        tm.assert_series_equal(series, expected)
+
     def test_iloc_setitem_list(self):
 
         # setitem with an iloc list