CODEWITHSUNDEEP
pythonpandas
Matt Savoie
Matt Savoie

Reputation: 772

How do I change order/grouping/level of pandas MultiIndex columns?

I'm trying to reorder/swaplevel/pivot/something columns in a pandas dataframe. The columns are a MultiIndex, but I can't find the sauce to do what I want.

The fastest varying column in my multiIndex is month, but I would like it to be the slowest varying column.

I've got a nbviewer notebook if you would like to try it out yourself: http://nbviewer.ipython.org/gist/flamingbear/4cfac24c80fe34a67474

What I have:

+-------------------------------------------------------------------+
|+-----+------+------+-----+------+-----+-----+------+-----+-----+  |
||     |weight             |extent            |rank                ||
|+-----+------+------+-----+------+-----+-----+------+-----+-----+  |
||month|'1Jan'|'Feb' |'Mar'|'1Jan'|'Feb'|'Mar'|'1Jan'|'Feb'|'Mar'|  |
|+-----+------+------+-----+------+-----+-----+------+-----+-----+  |
||year |      |      |     |      |     |     |      |     |     |  |
|+-----+------+------+-----+------+-----+-----+------+-----+-----+  |
||2000 |45.1  |46.1  |25.1 |13.442|14.94|15.02|13    |17   |14   |  |
|+-----+------+------+-----+------+-----+-----+------+-----+-----+  |
||2001 |85.0  |16.0  |49.0 |13.380|14.81|15.14|12    |15   |17   |  |
|+-----+------+------+-----+------+-----+-----+------+-----+-----+  |
||2002 |90.0  |33.0  |82.0 |13.590|15.13|14.88|15    |22   |10   |  |
|+-----+------+------+-----+------+-----+-----+------+-----+-----+  |
||2003 |47.0  |34.0  |78.0 |13.640|14.83|15.27|17    |16   |22   |  |
|+-----+------+------+-----+------+-----+-----+------+-----+-----+  |
+-------------------------------------------------------------------+

What I want

+------------------------------------------------------------------+
|+-----+------+------+----+------+------+-----+------+------+----+ |
||month|1Jan              |Feb                |Mar                ||
|+-----+------+------+----+------+------+-----+------+------+----+ |
||     |weight|extent|rank|weight|extent|rank |weight|extent|rank| |
|+-----+------+------+----+------+------+-----+------+------+----+ |
||year |      |      |    |      |      |     |      |      |    | |
|+-----+------+------+----+------+------+-----+------+------+----+ |
||2000 |45.1  |13.442|13  |46.1  |14.94 |17   | 25.1 |15.02 |14  | |
|+-----+------+------+----+------+------+-----+------+------+----+ |
||2001 |85.0  |13.380|12  |16.0  |14.81 |15   | 49.0 |15.14 |17  | |
|+-----+------+------+----+------+------+-----+------+------+----+ |
||2002 |90.0  |13.590|15  |33.0  |15.13 |22   | 82.0 |14.88 |10  | |
|+-----+------+------+----+------+------+-----+------+------+----+ |
||2003 |47.0  |13.640|17  |34.0  |14.83 |16   | 78.0 |15.27 |22  | |
|+-----+------+------+-----------+------+-----+------+------+----+ |
+------------------------------------------------------------------+

Any help would be appreciated. I can work with my original DataFrame, but writing to a CSV with the desired ordering would be fantastic.

Thanks in advance, Matt

Upvotes: 54

Views: 37932

Answers (3)

Antony Hatchkins
Antony Hatchkins

Reputation: 34014

Another method that does not need explicit index sorting is

df.stack(0).unstack()

Upvotes: 2

Alexander
Alexander

Reputation: 109686

Your columns are a MultiIndex. You need to reassign the DataFrame's columns with a new MultiIndex created from swapping levels of the existing one:

df.columns = df.columns.swaplevel(0, 1)
df.sort_index(axis=1, level=0, inplace=True)
>>> df

month   '1Jan'                 'Feb'                 'Mar'              
        weight  extent  rank  weight  extent  rank  weight  extent  rank
year                                                                    
2000      45.1  13.442    13    46.1   14.94    17    25.1   15.02    14
2001      85.0  13.380    12    16.0   14.81    15    49.0   15.14    17
2002      90.0  13.590    15    33.0   15.13    22    82.0   14.88    10
2003      47.0  13.640    17    34.0   14.83    16    78.0   15.27    22

You can then export to csv:

df.to_csv(filename)

Upvotes: 80

Mehdi S
Mehdi S

Reputation: 313

Since levels indices are no more mandatory you can have even more simple way to achieve the level swapping of multi-index dataframe:

df = df.swaplevel(axis='columns')

Upvotes: 9

Related Questions