Merge pull request #5227 from jreback/enlarge_empty

BUG: allow enlargement to work with empty objects (GH5226)

Author: jreback

doc/source/release.rst

@@ -206,7 +206,7 @@ API Changes
     (:issue:`4384`, :issue:`4375`, :issue:`4372`)
   - ``Series.get`` with negative indexers now returns the same as ``[]`` (:issue:`4390`)
   - allow ``ix/loc`` for Series/DataFrame/Panel to set on any axis even when the single-key is not currently contained in
-    the index for that axis (:issue:`2578`)
+    the index for that axis (:issue:`2578`, :issue:`5226`)
   - ``at`` now will enlarge the object inplace (and return the same) (:issue:`2578`)
 
   - ``HDFStore``

pandas/core/frame.py

@@ -1865,6 +1865,15 @@ def _setitem_frame(self, key, value):
 
         self.where(-key, value, inplace=True)
 
+    def _ensure_valid_index(self, value):
+        """
+        ensure that if we don't have an index, that we can create one from the passed value
+        """
+        if not len(self.index):
+            if not isinstance(value, Series):
+                raise ValueError("cannot set a frame with no defined index and a non-series")
+            self._data.set_axis(1, value.index.copy(), check_axis=False)
+
     def _set_item(self, key, value):
         """
         Add series to DataFrame in specified column.
@@ -1875,6 +1884,7 @@ def _set_item(self, key, value):
         Series/TimeSeries will be conformed to the DataFrame's index to
         ensure homogeneity.
         """
+        self._ensure_valid_index(value)
         value = self._sanitize_column(key, value)
         NDFrame._set_item(self, key, value)
 
@@ -1890,6 +1900,7 @@ def insert(self, loc, column, value, allow_duplicates=False):
         column : object
         value : int, Series, or array-like
         """
+        self._ensure_valid_index(value)
         value = self._sanitize_column(column, value)
         self._data.insert(
             loc, column, value, allow_duplicates=allow_duplicates)
@@ -1900,7 +1911,7 @@ def _sanitize_column(self, key, value):
         if _is_sequence(value):
             is_frame = isinstance(value, DataFrame)
             if isinstance(value, Series) or is_frame:
-                if value.index.equals(self.index):
+                if value.index.equals(self.index) or not len(self.index):
                     # copy the values
                     value = value.values.copy()
                 else:

pandas/core/indexing.py

@@ -173,9 +173,19 @@ def _setitem_with_indexer(self, indexer, value):
                     if self.ndim > 1 and i == self.obj._info_axis_number:
 
                         # add the new item, and set the value
-                        new_indexer = _convert_from_missing_indexer_tuple(indexer)
+                        # must have all defined axes if we have a scalar
+                        # or a list-like on the non-info axes if we have a list-like
+                        len_non_info_axes = [ len(_ax) for _i, _ax in enumerate(self.obj.axes) if _i != i ]
+                        if any([ not l for l in len_non_info_axes ]):
+                            if not is_list_like(value):
+                                raise ValueError("cannot set a frame with no defined index and a scalar")
+                            self.obj[key] = value
+                            return self.obj
+
                         self.obj[key] = np.nan
-                        self.obj.loc[new_indexer] = value
+
+                        new_indexer = _convert_from_missing_indexer_tuple(indexer, self.obj.axes)
+                        self._setitem_with_indexer(new_indexer, value)
                         return self.obj
 
                     # reindex the axis
@@ -208,12 +218,21 @@ def _setitem_with_indexer(self, indexer, value):
                     else:
                         new_index = _safe_append_to_index(index, indexer)
 
-                    new_values = np.concatenate([self.obj.values, [value]])
+                    # this preserves dtype of the value
+                    new_values = Series([value]).values
+                    if len(self.obj.values):
+                        new_values = np.concatenate([self.obj.values, new_values])
+
                     self.obj._data = self.obj._constructor(new_values, index=new_index, name=self.obj.name)
                     self.obj._maybe_update_cacher(clear=True)
                     return self.obj
 
                 elif self.ndim == 2:
+
+                    # no columns and scalar
+                    if not len(self.obj.columns):
+                        raise ValueError("cannot set a frame with no defined columns")
+
                     index = self.obj._get_axis(0)
                     labels = _safe_append_to_index(index, indexer)
                     self.obj._data = self.obj.reindex_axis(labels,0)._data
@@ -410,16 +429,17 @@ def _align_series(self, indexer, ser):
                         new_ix = Index([new_ix])
                     else:
                         new_ix = Index(new_ix.ravel())
-                    if ser.index.equals(new_ix):
+                    if ser.index.equals(new_ix) or not len(new_ix):
                         return ser.values.copy()
+
                     return ser.reindex(new_ix).values
 
                 # 2 dims
                 elif single_aligner and is_frame:
 
                     # reindex along index
                     ax = self.obj.axes[1]
-                    if ser.index.equals(ax):
+                    if ser.index.equals(ax) or not len(ax):
                         return ser.values.copy()
                     return ser.reindex(ax).values
 
@@ -819,6 +839,12 @@ def _convert_to_indexer(self, obj, axis=0, is_setter=False):
             # if we are setting and its not a valid location
             # its an insert which fails by definition
             if is_setter:
+
+                # always valid
+                if self.name == 'loc':
+                    return { 'key' : obj }
+
+                # a positional
                 if obj >= len(self.obj) and not isinstance(labels, MultiIndex):
                     raise ValueError("cannot set by positional indexing with enlargement")
 
@@ -1307,11 +1333,11 @@ def _convert_missing_indexer(indexer):
 
     return indexer, False
 
-def _convert_from_missing_indexer_tuple(indexer):
+def _convert_from_missing_indexer_tuple(indexer, axes):
     """ create a filtered indexer that doesn't have any missing indexers """
-    def get_indexer(_idx):
-        return _idx['key'] if isinstance(_idx,dict) else _idx
-    return tuple([ get_indexer(_idx) for _i, _idx in enumerate(indexer) ])
+    def get_indexer(_i, _idx):
+        return axes[_i].get_loc(_idx['key']) if isinstance(_idx,dict) else _idx
+    return tuple([ get_indexer(_i, _idx) for _i, _idx in enumerate(indexer) ])
 
 def _safe_append_to_index(index, key):
     """ a safe append to an index, if incorrect type, then catch and recreate """

pandas/tests/test_indexing.py

@@ -1542,6 +1542,56 @@ def f():
             df.ix[100,:] = df.ix[0]
         self.assertRaises(ValueError, f)
 
+    def test_partial_set_empty(self):
+
+        # GH5226
+
+        # partially set with an empty object
+        # series
+        s = Series()
+        s.loc[1] = 1
+        assert_series_equal(s,Series([1],index=[1]))
+        s.loc[3] = 3
+        assert_series_equal(s,Series([1,3],index=[1,3]))
+
+        s = Series()
+        s.loc[1] = 1.
+        assert_series_equal(s,Series([1.],index=[1]))
+        s.loc[3] = 3.
+        assert_series_equal(s,Series([1.,3.],index=[1,3]))
+
+        s = Series()
+        s.loc['foo'] = 1
+        assert_series_equal(s,Series([1],index=['foo']))
+        s.loc['bar'] = 3
+        assert_series_equal(s,Series([1,3],index=['foo','bar']))
+        s.loc[3] = 4
+        assert_series_equal(s,Series([1,3,4],index=['foo','bar',3]))
+
+        # partially set with an empty object
+        # frame
+        df = DataFrame()
+
+        def f():
+            df.loc[1] = 1
+        self.assertRaises(ValueError, f)
+        def f():
+            df.loc[1] = Series([1],index=['foo'])
+        self.assertRaises(ValueError, f)
+        def f():
+            df.loc[:,1] = 1
+        self.assertRaises(ValueError, f)
+
+        df2 = DataFrame()
+        df2[1] = Series([1],index=['foo'])
+        df.loc[:,1] = Series([1],index=['foo'])
+        assert_frame_equal(df,DataFrame([[1]],index=['foo'],columns=[1]))
+        assert_frame_equal(df,df2)
+
+        df = DataFrame(columns=['A','B'])
+        df.loc[3] = [6,7]
+        assert_frame_equal(df,DataFrame([[6,7]],index=[3],columns=['A','B']))
+
     def test_cache_updating(self):
         # GH 4939, make sure to update the cache on setitem