Merge pull request #7593 from jreback/timedelta_nat

BUG: Bug in timedelta inference when assigning an incomplete Series (GH7592)

Author: jreback

doc/source/v0.14.1.txt

@@ -172,7 +172,7 @@ Bug Fixes
 - Regression in datetimelike slice indexing with a duplicated index and non-exact end-points (:issue:`7523`)
 - Bug in setitem with list-of-lists and single vs mixed types (:issue:`7551`:)
 - Bug in timeops with non-aligned Series (:issue:`7500`)
-
+- Bug in timedelta inference when assigning an incomplete Series (:issue:`7592`)
 
 - Bug in ``value_counts`` where ``NaT`` did not qualify as missing (``NaN``) (:issue:`7423`)
 

pandas/core/common.py

@@ -2603,7 +2603,7 @@ def check_main():
 def in_qtconsole():
     """
     check if we're inside an IPython qtconsole
-    
+
     DEPRECATED: This is no longer needed, or working, in IPython 3 and above.
     """
     try:
@@ -2622,7 +2622,7 @@ def in_qtconsole():
 def in_ipnb():
     """
     check if we're inside an IPython Notebook
-    
+
     DEPRECATED: This is no longer used in pandas, and won't work in IPython 3
     and above.
     """

pandas/core/internals.py

@@ -1037,9 +1037,11 @@ class FloatBlock(FloatOrComplexBlock):
     def _can_hold_element(self, element):
         if is_list_like(element):
             element = np.array(element)
-            return issubclass(element.dtype.type, (np.floating, np.integer))
-        return (isinstance(element, (float, int, np.float_, np.int_)) and
-                not isinstance(bool, np.bool_))
+            tipo = element.dtype.type
+            return issubclass(tipo, (np.floating, np.integer)) and not issubclass(
+                tipo, (np.datetime64, np.timedelta64))
+        return isinstance(element, (float, int, np.float_, np.int_)) and not isinstance(
+            element, (bool, np.bool_, datetime, timedelta, np.datetime64, np.timedelta64))
 
     def _try_cast(self, element):
         try:
@@ -1099,7 +1101,8 @@ class IntBlock(NumericBlock):
     def _can_hold_element(self, element):
         if is_list_like(element):
             element = np.array(element)
-            return issubclass(element.dtype.type, np.integer)
+            tipo = element.dtype.type
+            return issubclass(tipo, np.integer) and not issubclass(tipo, (np.datetime64, np.timedelta64))
         return com.is_integer(element)
 
     def _try_cast(self, element):

pandas/tests/test_frame.py

@@ -32,7 +32,8 @@
 import pandas.core.format as fmt
 import pandas.core.datetools as datetools
 from pandas import (DataFrame, Index, Series, notnull, isnull,
-                    MultiIndex, DatetimeIndex, Timestamp, date_range, read_csv)
+                    MultiIndex, DatetimeIndex, Timestamp, date_range, read_csv,
+                    _np_version_under1p7)
 import pandas as pd
 from pandas.parser import CParserError
 from pandas.util.misc import is_little_endian
@@ -2180,11 +2181,11 @@ def test_set_index_cast_datetimeindex(self):
         # reset_index with single level
         for tz in ['UTC', 'Asia/Tokyo', 'US/Eastern']:
             idx = pd.date_range('1/1/2011', periods=5, freq='D', tz=tz, name='idx')
-            df = pd.DataFrame({'a': range(5), 'b': ['A', 'B', 'C', 'D', 'E']}, index=idx) 
+            df = pd.DataFrame({'a': range(5), 'b': ['A', 'B', 'C', 'D', 'E']}, index=idx)
 
             expected = pd.DataFrame({'idx': [datetime(2011, 1, 1), datetime(2011, 1, 2),
                                              datetime(2011, 1, 3), datetime(2011, 1, 4),
-                                             datetime(2011, 1, 5)], 
+                                             datetime(2011, 1, 5)],
                                      'a': range(5), 'b': ['A', 'B', 'C', 'D', 'E']},
                                      columns=['idx', 'a', 'b'])
             expected['idx'] = expected['idx'].apply(lambda d: pd.Timestamp(d, tz=tz))
@@ -3757,6 +3758,28 @@ def test_operators_timedelta64(self):
         self.assertTrue(df['off1'].dtype == 'timedelta64[ns]')
         self.assertTrue(df['off2'].dtype == 'timedelta64[ns]')
 
+    def test_datetimelike_setitem_with_inference(self):
+        if _np_version_under1p7:
+            raise nose.SkipTest("numpy < 1.7")
+
+        # GH 7592
+        # assignment of timedeltas with NaT
+
+        one_hour = timedelta(hours=1)
+        df = DataFrame(index=date_range('20130101',periods=4))
+        df['A'] = np.array([1*one_hour]*4, dtype='m8[ns]')
+        df.loc[:,'B'] = np.array([2*one_hour]*4, dtype='m8[ns]')
+        df.loc[:3,'C'] = np.array([3*one_hour]*3, dtype='m8[ns]')
+        df.ix[:,'D'] = np.array([4*one_hour]*4, dtype='m8[ns]')
+        df.ix[:3,'E'] = np.array([5*one_hour]*3, dtype='m8[ns]')
+        df['F'] = np.timedelta64('NaT')
+        df.ix[:-1,'F'] = np.array([6*one_hour]*3, dtype='m8[ns]')
+        df.ix[-3:,'G'] = date_range('20130101',periods=3)
+        df['H'] = np.datetime64('NaT')
+        result = df.dtypes
+        expected = Series([np.dtype('timedelta64[ns]')]*6+[np.dtype('datetime64[ns]')]*2,index=list('ABCDEFGH'))
+        assert_series_equal(result,expected)
+
     def test_new_empty_index(self):
         df1 = DataFrame(randn(0, 3))
         df2 = DataFrame(randn(0, 3))