在控制MultiIndex值的同时串联熊猫数据框和序列

Question

Pandas concat method allows you to concatenate mixtures of Series and Dataframes but the default way it infers column names for the series in the resulting dataframe is not quite what I want.

Example:

Say I have a dictionary containing a collection of dataframes and series as values.

dict_of_series_and_dataframes = {
    'x': pd.Series([1, 2, 3]),
    'y': pd.Series([6, 5, 4]),
    'sizes': pd.DataFrame(100, columns=[1, 2, 3], index=range(3)),
    'z': pd.Series([0.1, 0.2, 0.3])
}

Combining them into one dataframe is very easy:

>>> pd.concat(dict_of_series_and_frames, axis=1)
  sizes            x  y    z
      1    2    3  0  1    2
0   100  100  100  1  6  0.1
1   100  100  100  2  5  0.2
2   100  100  100  3  4  0.3

The problem is the way Pandas has filled in the second level for the series. Seems to be a simple integer count (0, 1, 2, ...). I would like to do something more logical such as labelling all the series ('Series name', None) or grouping them as ('Other', 'Series name') to make them easier to index later (all my series have unique names).

I found out that pandas puts the series name in the second level if it has one:

dict_of_series_and_frames = {
    'x': pd.Series([1, 2, 3], name='x'),
    'y': pd.Series([6, 5, 4], name='y'),
    'sizes': pd.DataFrame(100, columns=[1, 2, 3], index=range(3)),
    'z': pd.Series([0.1, 0.2, 0.3])
}

>>> pd.concat(dict_of_series_and_frames, axis=1)
  sizes            x  y    z
      1    2    3  x  y    0
0   100  100  100  1  6  0.1
1   100  100  100  2  5  0.2
2   100  100  100  3  4  0.3

But I don't want to have to ensure that all the series are named correctly.

Other than ignoring pandas attempt to build the index and doing it all by hand...

>>> col_names = []
>>> for key, values in dict_of_series_and_frames.items():
...     try:
...         for value in values.columns:
...             col_names.append((key, value))
...     except AttributeError:
...         col_names.append((key, '-'))  # or ('Other', key) say
... 
>>> df = pd.concat(dict_of_series_and_frames, 
...                keys=dict_of_series_and_frames.keys(), 
...                axis=1, ignore_index=True)
>>> df.columns = pd.MultiIndex.from_tuples(col_names)
>>> df
   x  y sizes              z
   -  -     1    2    3    -
0  1  6   100  100  100  0.1
1  2  5   100  100  100  0.2
2  3  4   100  100  100  0.3

am I missing a simpler way to get the desired result above or something similar?

Ideally in one line using concat.

Answer 1

You can modify/add series names already in the dictionary, and then apply concatenation:

for k,v in dict_of_series_and_frames.items(): 
         if isinstance(v,pd.core.series.Series): 
             v.name="-"