BUG: provide for automatic conversion of object -> datetime64[ns] typ…

…es upon creation (in make_block)

     this obviates the need to convert_objects (mostly)
     in addition, enabled setting of NaT in datetime64[ns] columns via np.nan (on-the-fly-conversion)

pandas/core/indexing.py

@@ -113,11 +113,25 @@ def _setitem_with_indexer(self, indexer, value):
                     if np.prod(values.shape):
                         values[plane_indexer] = value
             except ValueError:
-                for item, v in zip(item_labels[het_idx], value):
-                    data = self.obj[item]
-                    values = data.values
-                    if np.prod(values.shape):
-                        values[plane_indexer] = v
+
+                # convert nan to iNaT if possible
+                if data.dtype == 'M8[ns]':
+                    mask = com._isnull(value)
+                    if np.isscalar(value) and mask:
+                        from pandas import tslib
+                        value = tslib.iNaT
+                        values[plane_indexer] = value
+                    elif isinstance(value, np.array) and mask.any():
+                        from pandas import tslib
+                        value = value.copy()
+                        value.putmask(iNat,mask)
+                        values[plane_indexer] = value
+                else:
+                    for item, v in zip(item_labels[het_idx], value):
+                        data = self.obj[item]
+                        values = data.values
+                        if np.prod(values.shape):
+                            values[plane_indexer] = v
         else:
             if isinstance(indexer, tuple):
                 indexer = _maybe_convert_ix(*indexer)

pandas/core/internals.py

@@ -446,6 +446,7 @@ def get_values(self, dtype):
 def make_block(values, items, ref_items):
     dtype = values.dtype
     vtype = dtype.type
+    klass = None
 
     if issubclass(vtype, np.floating):
         klass = FloatBlock
@@ -459,7 +460,20 @@ def make_block(values, items, ref_items):
         klass = IntBlock
     elif dtype == np.bool_:
         klass = BoolBlock
-    else:
+
+    # try to infer a datetimeblock
+    if klass is None and np.prod(values.shape):
+        inferred_type = lib.infer_dtype(values.flatten())
+        if inferred_type == 'datetime':
+
+            # we have an object array that has been inferred as datetime, so convert it
+            try:
+                values = tslib.array_to_datetime(values.flatten()).reshape(values.shape)
+                klass = DatetimeBlock
+            except: # it already object, so leave it
+                pass
+
+    if klass is None:
         klass = ObjectBlock
 
     return klass(values, items, ref_items, ndim=values.ndim)

pandas/tests/test_frame.py

@@ -21,7 +21,7 @@
 import pandas.core.format as fmt
 import pandas.core.datetools as datetools
 from pandas.core.api import (DataFrame, Index, Series, notnull, isnull,
-                             MultiIndex, DatetimeIndex)
+                             MultiIndex, DatetimeIndex, Timestamp)
 from pandas.io.parsers import read_csv
 
 from pandas.util.testing import (assert_almost_equal,
@@ -1073,6 +1073,27 @@ def test_setitem_single_column_mixed(self):
         expected = [nan, 'qux', nan, 'qux', nan]
         assert_almost_equal(df['str'].values, expected)
 
+    def test_setitem_single_column_mixed_datetime(self):
+        df = DataFrame(randn(5, 3), index=['a', 'b', 'c', 'd', 'e'],
+                       columns=['foo', 'bar', 'baz'])
+
+        df['timestamp'] = Timestamp('20010102')
+
+        # check our dtypes
+        result = df.get_dtype_counts()
+        expected = Series({ 'float64' : 3, 'datetime64[ns]' : 1})
+        assert_series_equal(result, expected)
+
+        # set an allowable datetime64 type
+        from pandas import tslib
+        df.ix['b','timestamp'] = tslib.iNaT
+
+        # this fails because nan is a float type
+        df.ix['b','timestamp'] = nan
+
+        # prior to 0.10.1 this failed
+        #self.assertRaises(TypeError, df.ix.__setitem__, ('c','timestamp'), nan)
+
     def test_setitem_frame(self):
         piece = self.frame.ix[:2, ['A', 'B']]
         self.frame.ix[-2:, ['A', 'B']] = piece.values

pandas/tseries/tests/test_timeseries.py

@@ -1244,9 +1244,9 @@ def test_append_concat(self):
 
     def test_set_dataframe_column_ns_dtype(self):
         x = DataFrame([datetime.now(), datetime.now()])
-        self.assert_(x[0].dtype == object)
+        #self.assert_(x[0].dtype == object)
 
-        x[0] = to_datetime(x[0])
+        #x[0] = to_datetime(x[0])
         self.assert_(x[0].dtype == np.dtype('M8[ns]'))
 
     def test_groupby_count_dateparseerror(self):
@@ -2075,6 +2075,7 @@ def test_get_level_values_box(self):
     def test_frame_apply_dont_convert_datetime64(self):
         from pandas.tseries.offsets import BDay
         df = DataFrame({'x1': [datetime(1996,1,1)]})
+
         df = df.applymap(lambda x: x+BDay())
         df = df.applymap(lambda x: x+BDay())