GH 9016: Bitwise operation weirdness

doc/source/whatsnew/v0.16.0.txt

@@ -88,6 +88,35 @@ Backwards incompatible API changes
 The prior style can be achieved with matplotlib's `axhline` or `axvline`
 methods (:issue:`9088`).
 
+
+- ``Series`` now supports bitwise operation for integral types (:issue:`9016`)
+
+  Previously even if the input dtypes where integral, the output dtype was coerced to bool.
+
+  .. code-block:: python
+    In [2]: pd.Series([0,1,2,3], list('abcd')) | pd.Series([4,4,4,4], list('abcd'))
+    Out[2]:
+    a    True
+    b    True
+    c    True
+    d    True
+    dtype: bool
+
+  Now if the input dtypes are integral, the output dtype is also integral and the output
+  values are the result of the bitwise operation.
+
+  .. code-block:: python
+
+    In [2]: pd.Series([0,1,2,3], list('abcd')) | pd.Series([4,4,4,4], list('abcd'))
+    Out[2]:
+    a    4
+    b    5
+    c    6
+    d    7
+    dtype: int64
+
+
+
 Deprecations
 ~~~~~~~~~~~~
 

pandas/core/ops.py

@@ -654,20 +654,31 @@ def na_op(x, y):
         return result
 
     def wrapper(self, other):
+        is_self_int_dtype = com.is_integer_dtype(self.dtype)
+
+        fill_int = lambda x: x.fillna(0)
+        fill_bool = lambda x: x.fillna(False).astype(bool)
+
         if isinstance(other, pd.Series):
             name = _maybe_match_name(self, other)
+            other = other.reindex_like(self)
+            is_other_int_dtype = com.is_integer_dtype(other.dtype)
+            other = fill_int(other) if is_other_int_dtype else fill_bool(other)
 
-            other = other.reindex_like(self).fillna(False).astype(bool)
-            return self._constructor(na_op(self.values, other.values),
+            filler = fill_int if is_self_int_dtype and is_other_int_dtype else fill_bool
+            return filler(self._constructor(na_op(self.values, other.values),
                                      index=self.index,
-                                     name=name).fillna(False).astype(bool)
+                                     name=name))
+
         elif isinstance(other, pd.DataFrame):
             return NotImplemented
+
         else:
-            # scalars
-            res = self._constructor(na_op(self.values, other),
-                                    index=self.index).fillna(False)
-            return res.astype(bool).__finalize__(self)
+            # scalars, list, tuple, np.array
+            filler = fill_int if is_self_int_dtype and com.is_integer_dtype(np.asarray(other)) else fill_bool
+            return filler(self._constructor(na_op(self.values, other),
+                                    index=self.index)).__finalize__(self)
+
     return wrapper
 
 

pandas/tests/test_series.py

@@ -3790,6 +3790,96 @@ def test_comparison_label_based(self):
         for v in [np.nan]:
             self.assertRaises(TypeError, lambda : t & v)
 
+    def test_operators_bitwise(self):
+        # GH 9016: support bitwise op for integer types
+        index = list('bca')
+
+        s_tft = Series([True, False, True], index=index)
+        s_fff = Series([False, False, False], index=index)
+        s_tff = Series([True, False, False], index=index)
+        s_empty = Series([])
+        s_0101 = Series([0,1,0,1])
+        s_0123 = Series(range(4))
+        s_3333 = Series([3] * 4)
+        s_4444 = Series([4] * 4)
+
+        res = s_tft & s_empty
+        expected = s_fff
+        assert_series_equal(res, expected)
+
+        res = s_tft | s_empty
+        expected = s_tft
+        assert_series_equal(res, expected)
+
+        res = s_0123 & s_3333
+        expected = Series(range(4))
+        assert_series_equal(res, expected)
+
+        res = s_0123 | s_4444
+        expected = Series(range(4, 8))
+        assert_series_equal(res, expected)
+
+        s_a0b1c0 = Series([1], list('b'))
+
+        res = s_tft & s_a0b1c0
+        expected = s_tff
+        assert_series_equal(res, expected)
+
+        res = s_tft | s_a0b1c0
+        expected = s_tft
+        assert_series_equal(res, expected)
+
+        n0 = 0
+        res = s_tft & n0
+        expected = s_fff
+        assert_series_equal(res, expected)
+
+        res = s_0123 & n0
+        expected = Series([0] * 4)
+        assert_series_equal(res, expected)
+
+        n1 = 1
+        res = s_tft & n1
+        expected = s_tft
+        assert_series_equal(res, expected)
+
+        res = s_0123 & n1
+        expected = Series([0, 1, 0, 1])
+        assert_series_equal(res, expected)
+
+        s_1111 = Series([1]*4, dtype='int8')
+        res = s_0123 & s_1111
+        expected = Series([0, 1, 0, 1], dtype='int64')
+        assert_series_equal(res, expected)
+
+        res = s_0123.astype(np.int16) | s_1111.astype(np.int32)
+        expected = Series([1, 1, 3, 3], dtype='int32')
+        assert_series_equal(res, expected)
+
+        self.assertRaises(TypeError, lambda: s_1111 & 'a')
+        self.assertRaises(TypeError, lambda: s_1111 & ['a','b','c','d'])
+        self.assertRaises(TypeError, lambda: s_0123 & np.NaN)
+        self.assertRaises(TypeError, lambda: s_0123 & 3.14)
+        self.assertRaises(TypeError, lambda: s_0123 & [0.1, 4, 3.14, 2])
+
+        # s_0123 will be all false now because of reindexing like s_tft
+        assert_series_equal(s_tft & s_0123, Series([False] * 3, list('bca')))
+        # s_tft will be all false now because of reindexing like s_0123
+        assert_series_equal(s_0123 & s_tft, Series([False] * 4))
+        assert_series_equal(s_0123 & False, Series([False] * 4))
+        assert_series_equal(s_0123 ^ False, Series([False, True, True, True]))
+        assert_series_equal(s_0123 & [False], Series([False] * 4))
+        assert_series_equal(s_0123 & (False), Series([False] * 4))
+        assert_series_equal(s_0123 & Series([False, np.NaN, False, False]), Series([False] * 4))
+
+        s_ftft = Series([False, True, False, True])
+        assert_series_equal(s_0123 & Series([0.1, 4, -3.14, 2]), s_ftft)
+
+        s_abNd = Series(['a','b',np.NaN,'d'])
+        res = s_0123 & s_abNd
+        expected = s_ftft
+        assert_series_equal(res, expected)
+
     def test_between(self):
         s = Series(bdate_range('1/1/2000', periods=20).asobject)
         s[::2] = np.nan