COMPAT: map infers all-nan / empty correctly (pandas-dev#18491)

Author: jreback

doc/source/whatsnew/v0.22.0.txt

@@ -75,7 +75,7 @@ Other API Changes
 - :class:`CacheableOffset` and :class:`WeekDay` are no longer available in the ``pandas.tseries.offsets`` module (:issue:`17830`)
 - `tseries.frequencies.get_freq_group()` and `tseries.frequencies.DAYS` are removed from the public API (:issue:`18034`)
 - :func:`Series.truncate` and :func:`DataFrame.truncate` will raise a ``ValueError`` if the index is not sorted instead of an unhelpful ``KeyError`` (:issue:`17935`)
-- :func:`Index.map` can now accept ``Series`` and dictionary input objects (:issue:`12756`).
+- :func:`Index.map` can now accept ``Series`` and dictionary input objects (:issue:`12756`, :issue:`18482`).
 - :func:`Dataframe.unstack` will now default to filling with ``np.nan`` for ``object`` columns. (:issue:`12815`)
 - :class:`IntervalIndex` constructor will raise if the ``closed`` parameter conflicts with how the input data is inferred to be closed (:issue:`18421`)
 - Inserting missing values into indexes will work for all types of indexes and automatically insert the correct type of missing value (``NaN``, ``NaT``, etc.) regardless of the type passed in (:issue:`18295`)

pandas/core/indexes/base.py

@@ -2910,7 +2910,10 @@ def map(self, mapper, na_action=None):
         from .multi import MultiIndex
         new_values = super(Index, self)._map_values(
             mapper, na_action=na_action)
+
         attributes = self._get_attributes_dict()
+
+        # we can return a MultiIndex
         if new_values.size and isinstance(new_values[0], tuple):
             if isinstance(self, MultiIndex):
                 names = self.names
@@ -2923,8 +2926,25 @@ def map(self, mapper, na_action=None):
 
         attributes['copy'] = False
 
-        # we infer the result types based on the
-        # returned values
+        # we want to try to return our original dtype
+        # ints infer to integer, but if we have
+        # uints, would prefer to return these
+        if is_unsigned_integer_dtype(self.dtype):
+            inferred = lib.infer_dtype(new_values)
+            if inferred == 'integer':
+                attributes['dtype'] = self.dtype
+
+        elif not new_values.size:
+            # empty
+            attributes['dtype'] = self.dtype
+        elif isna(new_values).all():
+            # all nan
+            inferred = lib.infer_dtype(self)
+            if inferred in ['datetime', 'datetime64',
+                            'timedelta', 'timedelta64',
+                            'period']:
+                new_values = [libts.NaT] * len(new_values)
+
         return Index(new_values, **attributes)
 
     def isin(self, values, level=None):

pandas/tests/indexes/common.py

@@ -1007,31 +1007,33 @@ def test_searchsorted_monotonic(self, indices):
                 indices._searchsorted_monotonic(value, side='left')
 
     def test_map(self):
+        # callable
         index = self.create_index()
+        expected = index
+        result = index.map(lambda x: x)
+        tm.assert_index_equal(result, expected)
 
-        # From output of UInt64Index mapping can't infer that we
-        #   shouldn't default to Int64
-        if isinstance(index, UInt64Index):
-            expected = Index(index.values.tolist())
-        else:
-            expected = index
+    @pytest.mark.parametrize(
+        "mapper",
+        [
+            lambda values, index: {i: e for e, i in zip(values, index)},
+            lambda values, index: pd.Series(values, index)])
+    def test_map_dictlike(self, mapper):
 
-        tm.assert_index_equal(index.map(lambda x: x), expected)
+        index = self.create_index()
+        if isinstance(index, (pd.CategoricalIndex, pd.IntervalIndex)):
+            pytest.skip("skipping tests for {}".format(type(index)))
 
-        identity_dict = {x: x for x in index}
-        tm.assert_index_equal(index.map(identity_dict), expected)
+        expected = index
 
-        # Use values to work around MultiIndex instantiation of series
-        identity_series = Series(expected.values, index=index)
-        tm.assert_index_equal(index.map(identity_series), expected)
+        identity = mapper(index.values, index)
+        result = index.map(identity)
+        tm.assert_index_equal(result, expected)
 
         # empty mappable
-        nan_index = pd.Index([np.nan] * len(index))
-        series_map = pd.Series()
-        tm.assert_index_equal(index.map(series_map), nan_index)
-
-        dict_map = {}
-        tm.assert_index_equal(index.map(dict_map), nan_index)
+        expected = pd.Index([np.nan] * len(index))
+        result = index.map(mapper(expected, index))
+        tm.assert_index_equal(result, expected)
 
     def test_putmask_with_wrong_mask(self):
         # GH18368

pandas/tests/indexes/datetimelike.py

@@ -1,7 +1,6 @@
 """ generic datetimelike tests """
 import pytest
 import pandas as pd
-import numpy as np
 from .common import Base
 import pandas.util.testing as tm
 
@@ -73,6 +72,6 @@ def test_map_dictlike(self, mapper):
 
         # empty map; these map to np.nan because we cannot know
         # to re-infer things
-        expected = pd.Index([np.nan] * len(self.index))
+        expected = pd.Index([pd.NaT] * len(self.index))
         result = self.index.map(mapper([], []))
         tm.assert_index_equal(result, expected)

pandas/tests/indexes/test_base.py

@@ -852,11 +852,15 @@ def test_map_tseries_indices_return_index(self):
         exp = Index(range(24), name='hourly')
         tm.assert_index_equal(exp, date_index.map(lambda x: x.hour))
 
-    def test_map_with_dict_and_series(self):
+    @pytest.mark.parametrize(
+        "mapper",
+        [
+            lambda values, index: {i: e for e, i in zip(values, index)},
+            lambda values, index: pd.Series(values, index)])
+    def test_map_dictlike(self, mapper):
         # GH 12756
         expected = Index(['foo', 'bar', 'baz'])
-        mapper = Series(expected.values, index=[0, 1, 2])
-        result = tm.makeIntIndex(3).map(mapper)
+        result = tm.makeIntIndex(3).map(mapper(expected.values, [0, 1, 2]))
         tm.assert_index_equal(result, expected)
 
         for name in self.indices.keys():
@@ -867,21 +871,16 @@ def test_map_with_dict_and_series(self):
                 # Cannot map duplicated index
                 continue
 
-            cur_index = self.indices[name]
-            expected = Index(np.arange(len(cur_index), 0, -1))
-            mapper = pd.Series(expected, index=cur_index)
-            result = cur_index.map(mapper)
-
-            tm.assert_index_equal(result, expected)
+            index = self.indices[name]
+            expected = Index(np.arange(len(index), 0, -1))
 
-            # If the mapper is empty the expected index type is Int64Index
-            # but the output defaults to Float64 so I treat it independently
-            mapper = {o: n for o, n in
-                      zip(cur_index, expected)}
+            # to match proper result coercion for uints
+            if name == 'uintIndex':
+                expected = expected.astype('uint64')
+            elif name == 'empty':
+                expected = Index([])
 
-            result = cur_index.map(mapper)
-            if not mapper:
-                expected = Float64Index([])
+            result = index.map(mapper(expected, index))
             tm.assert_index_equal(result, expected)
 
     def test_map_with_non_function_missing_values(self):

pandas/tests/indexes/test_interval.py

@@ -584,10 +584,6 @@ def test_repr_max_seq_item_setting(self):
     def test_repr_roundtrip(self):
         super(TestIntervalIndex, self).test_repr_roundtrip()
 
-    @pytest.mark.xfail(reason='get_indexer behavior does not currently work')
-    def test_map(self):
-        super(TestIntervalIndex, self).test_map()
-
     def test_get_item(self, closed):
         i = IntervalIndex.from_arrays((0, 1, np.nan), (1, 2, np.nan),
                                       closed=closed)