BUG: fix initialization of Series with dict containing NaN as key

closes pandas-dev#18480

Author: toobaz

doc/source/whatsnew/v0.22.0.txt

@@ -208,4 +208,5 @@ Other
 
 - Improved error message when attempting to use a Python keyword as an identifier in a numexpr query (:issue:`18221`)
 - Fixed a bug where creating a Series from an array that contains both tz-naive and tz-aware values will result in a Series whose dtype is tz-aware instead of object (:issue:`16406`)
+- Fixed construction of :class:`Series` from ``dict`` containing ``NaN`` as key (:issue:`18480`)
 -

pandas/core/base.py

@@ -859,7 +859,6 @@ def _map_values(self, mapper, na_action=None):
             a MultiIndex will be returned.
 
         """
-
         # we can fastpath dict/Series to an efficient map
         # as we know that we are not going to have to yield
         # python types
@@ -875,7 +874,7 @@ def _map_values(self, mapper, na_action=None):
                 # we specify the keys here to handle the
                 # possibility that they are tuples
                 from pandas import Series, Index
-                index = Index(mapper, tupleize_cols=False)
+                index = Index(mapper, tupleize_cols=True)
                 mapper = Series(mapper, index=index)
 
         if isinstance(mapper, ABCSeries):

pandas/core/series.py

@@ -198,18 +198,9 @@ def __init__(self, data=None, index=None, dtype=None, name=None,
                     data = data.reindex(index, copy=copy)
                 data = data._data
             elif isinstance(data, dict):
-                if index is None:
-                    if isinstance(data, OrderedDict):
-                        index = Index(data)
-                    else:
-                        index = Index(_try_sort(data))
-
-                try:
-                    data = index._get_values_from_dict(data)
-                except TypeError:
-                    data = ([data.get(i, np.nan) for i in index]
-                            if data else np.nan)
-
+                data, index = self._init_from_dict(data, index, dtype)
+                dtype = None
+                copy = False
             elif isinstance(data, SingleBlockManager):
                 if index is None:
                     index = data.index
@@ -303,6 +294,23 @@ def _can_hold_na(self):
 
     _index = None
 
+    def _init_from_dict(self, data, index, dtype):
+        # Looking for NaN in dict doesn't work ({np.nan : 1}[float('nan')]
+        # raises KeyError), so we iterate the entire dict, and align
+        if data:
+            keys, values = zip(*compat.iteritems(data))
+        else:
+            keys, values = [], []
+        s = Series(values, index=keys, dtype=dtype)
+        if index is not None and not index.identical(keys):
+            s = s.reindex(index)
+        elif not isinstance(data, OrderedDict):
+            try:
+                s = s.sort_index()
+            except TypeError:
+                pass
+        return s._data, s.index
+
     def _set_axis(self, axis, labels, fastpath=False):
         """ override generic, we want to set the _typ here """
 

pandas/tests/series/test_combine_concat.py

@@ -181,7 +181,7 @@ def test_concat_empty_series_dtypes(self):
         # categorical
         assert pd.concat([Series(dtype='category'),
                           Series(dtype='category')]).dtype == 'category'
-        assert pd.concat([Series(dtype='category'),
+        assert pd.concat([Series(np.array([]), dtype='category'),
                           Series(dtype='float64')]).dtype == 'float64'
         assert pd.concat([Series(dtype='category'),
                           Series(dtype='object')]).dtype == 'object'

pandas/tests/series/test_constructors.py

@@ -625,6 +625,32 @@ def test_constructor_dict(self):
         expected.iloc[1] = 1
         assert_series_equal(result, expected)
 
+    def test_constructor_dict_nan_key(self):
+        # GH 18480
+        d = {1: 'a', 2: 'b', np.nan: 'c'}
+        result = Series(d).sort_index()
+        expected = Series(['a', 'b', 'c'], index=[1, 2, np.nan])
+        assert_series_equal(result, expected)
+
+        # Different NaNs:
+        d = {1: 'a', 2: 'b', float('nan'): 'c', float('nan'): 'd'}
+        result = Series(d).sort_values()
+        expected = Series(['a', 'b', 'c', 'd'], index=[1, 2, np.nan, np.nan])
+        assert_series_equal(result, expected)
+
+        # NaN vs. None:
+        d = {1: 'a', 2: 'b', np.nan: 'c', None: 'd'}
+        result = Series(d).sort_values()
+        expected = Series(['a', 'b', 'c', 'd'], index=[1, 2, np.nan, None])
+        assert_series_equal(result, expected)
+
+        # MultiIndex:
+        d = {(1, 1): 'a', (2, 2): 'b', (3, np.nan): 'c'}
+        result = Series(d).sort_values()
+        expected = Series(['a', 'b', 'c'],
+                          index=Index([(1, 1), (2, 2), (3, np.nan)]))
+        assert_series_equal(result, expected)
+
     def test_constructor_dict_datetime64_index(self):
         # GH 9456
 
@@ -658,8 +684,6 @@ def test_constructor_tuple_of_tuples(self):
         s = Series(data)
         assert tuple(s) == data
 
-    @pytest.mark.xfail(reason='GH 18480 (Series initialization from dict with '
-                              'NaN keys')
     def test_constructor_dict_of_tuples(self):
         data = {(1, 2): 3,
                 (None, 5): 6}