将函数应用于分组的DataFrame后，Pandas sort_index给出奇怪的结果

Question

Basic setup:

I have a DataFrame with a MultiIndex on both the rows and the columns. The second level of the column index has float s for values.

I want to perform a groupby operation (grouping by the first level of the row index). The operation will add a few columns (also with float s as their labels) to each group and then return the group.

When I get the result back from my groupby operation, I can't seem to get the columns to sort properly.

Working example. First, set things up:

import pandas as pd
import numpy as np

np.random.seed(0)

col_level_1 = ['red', 'blue']
col_level_2 = [1., 2., 3., 4.]

row_level_1 = ['a', 'b']
row_level_2 = ['one', 'two']

col_idx = pd.MultiIndex.from_product([col_level_1, col_level_2], names=['color', 'numeral'])
row_idx = pd.MultiIndex.from_product([row_level_1, row_level_2], names=['letter', 'number'])

df = pd.DataFrame(np.random.randn(len(row_idx), len(col_idx)), index=row_idx, columns=col_idx)

Gives this DataFrame in df :

Then define my group operation and apply it:

def mygrpfun(group):
    for f in [1.5, 2.5, 3.5]:
        group[('red', f)] = 'hello'
        group[('blue', f)] = 'world'
    return group

result = df.groupby(level='letter').apply(mygrpfun).sort_index(axis=1)

Displaying result gives:

What's going on here? Why doesn't the 2nd level of the column index display in ascending order?

EDIT: In terms of context:

pd.__version__
Out[28]:
'0.14.0'
In [29]:

np.__version__
Out[29]:
'1.8.1'

Any help much appreciated.

Answer 1

The returned result looks as expected. You added columns. There was no guarantee that order imposed on those columns.

You could just reimpose ordering:

result = result[sorted(result.columns)]