熊猫数据透视表中的Multiindex

Question

I am working on a pivot table that looks like this:

            Style  Site AVS  End Qty.                                          \
JP SIZE                           116  120  140  ADULTS  L  M  O  OSFA  S  XL   
0        50935801  2664   0         0    0    0       0  0  0  0     0  0   3   
1        50935801  2807   0         0    0    0       0  0  0  0     0  0   3   
2        50935801  2832   0         0    0    0       0  0  0  0     0  0   3   
3        50935802  2702   1         0    0    0       0  0  1  0     0  0   0   
4        50985101  2849   0         0    0    0       0  0  3  0     0  0   0   

            Sales Qty.                              
JP SIZE  Total         116  120  140  ADULTS  L  M      
0            3           0    0    0       0  0  0 ...  
1            3           0    0    0       0  0  0 ...  
2            3           0    0    0       0  0  0 ...  
3            1           0    0    0       0  0 -1 ...  
4            3           0    0    0       0  0  0 ...  

And I would like to have only one vector of column headers that would be [Style, Site, AVS, 116, 120 , ... , Total , Sales Qty.]

But for the "Sales Qty." column, instead of the table that is there at the moment I would like only the total column (I can access it with jj['Sales Qty']['Total'] for the moment, so I guess I could save it in another variable, delete it and add it in the end)

Everything that I have tried so far has failed, I think it is because I don't understand very well how MultiIndex work yet.

Thanks in advance for any help you can provide on that!

Answer 1

There might be something more clever built in but one way is to work with MultiIndex as a list of tuples, and map out the new column names as you described.

def custom_rename(lvl1, lvl2):
    if lvl1 == 'End Qty.':
        return lvl2
    elif lvl1 == 'Sales Qty.' and lvl2 == 'Total':
        return 'Sales Qty.'
    elif lvl2 == '':
        return lvl1
    else:
        return '_'

Then apply to the columns and assign:

df.columns = [custom_rename(lvl1, lvl2) for lvl1, lvl2 in df.columns]

'_' above was used a marker for the columns no longer desired, so the last step would be to drop those.

df = df.drop('_', axis=1)