Add column to empty DataFrame?

[klonuo]

I'm reading 0.9 PDF documentation, about DataFrame object.

I make empty DataFrame:

>>> from pandas import DataFrame
>>> df = DataFrame()
>>> df
Empty DataFrame
Columns: array([], dtype=object)
Index: array([], dtype=object)
>>> df['A'] = [1, 2, 3]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Python27\lib\site-packages\pandas\core\frame.py", line 1804, in __setitem__
    self._set_item(key, value)
  File "C:\Python27\lib\site-packages\pandas\core\frame.py", line 1843, in _set_item
    value = self._sanitize_column(key, value)
  File "C:\Python27\lib\site-packages\pandas\core\frame.py", line 1872, in _sanitize_column
    assert(len(value) == len(self.index))
AssertionError

Error w/o any description.

If df has already columns, then I can add another:

>>> from pandas import DataFrame
>>> df = DataFrame.from_items([('A', [1, 2, 3])])
>>> df['B'] = [3, 2, 1]
>>> df
   A  B
0  1  3
1  2  2
2  3  1

I'm not sure if this is some kind of feature or it is a bug.

[lodagro]

If you use the pattern df[<column_label>] = <my_list>, it is required that len(df.index) == len(<my_list>).
This is somehow shown in the backtrace. What counts is that df has an index of the same length as my_list, other columns are not required.

In [35]: df = DataFrame(index=range(3))

In [36]: df
Out[36]: 
Empty DataFrame
Columns: array([], dtype=object)
Index: array([0, 1, 2], dtype=int64)

In [37]: df['A'] = [1, 2, 3]

In [38]: df
Out[38]: 
   A
0  1
1  2
2  3

[klonuo]

Thanks for explanation and your tip with indexing first

Maybe it's just in Python's philosophy things to be like this, although I expected different outcome. For example in VBS I don't need to initialize variable to be able to do operation on it, as it's assumed to be 0 or empty or else (depending on context) if not set. I can just type j=j+1, and it would work, j would be 1 (which is handy in loops). In Python I would have to declare j=0 first. Similar with lists.

OK, I think that adding column to empty DataFrame perhaps could be improved by taking the index from added column length, but maybe it doesn't fit in global scheme.

Is there other way of adding columns in DataFrame except df[<column_label>] = <my_list> that would work without initialization with index?

[changhiskhan]

I added a slightly more detailed error msg for this case.

I don't think it's a good idea to reindex input columns to index in non-empty cases and reindex the DataFrame in empty-cases, but maybe @wesm or @lodagro might have other opinions?

[lodagro]

Not sure if i understand your question. Do you propose when a columns is added to an empty/index-less DataFrame, to set a default index (for non Series) or re-use the index in case of a Series?