Pandas pivot table Percent Calculations

Question

Given the following data frame and pivot table:

import pandas as pd
df=pd.DataFrame({'A':['x','y','z','x','y','z'],
                 'B':['one','one','one','two','two','two'],
                 'C':[2,18,2,8,2,18]})
df

    A   B       C
0   x   one     2
1   y   one     18
2   z   one     2
3   x   two     8
4   y   two     2
5   z   two     18

table = pd.pivot_table(df, index=['A', 'B'],aggfunc=np.sum)

            C
A   B   
x   one     2
    two     8
y   one     18
    two     2
z   one     2
    two     18

I'd like to add 2 columns to this pivot table; one showing the percent of all values and another for percent within column A like this:

           C    % of Total  % of B
A   B
x   one    2    4%          20%
    two    8    16%         80%
y   one   18    36%         90%
    two    2    4%          10%
z   one    2    4%          10%
    two   18    36%         90%

Extra Credit:

I'd like a bottom summary row which has the sum of column C (it's okay if it also has 100% for the next 2 columns, but nothing is needed for those).

Answer 1

You can use:

table['% of Total'] = (table.C / table.C.sum() * 100).astype(str) + '%'
table['% of B'] = (table.C / table.groupby(level=0).C.transform(sum) * 100).astype(str) + '%'
print table
        C % of Total % of B
A B                        
x one   2       4.0%  20.0%
  two   8      16.0%  80.0%
y one  18      36.0%  90.0%
  two   2       4.0%  10.0%
z one   2       4.0%  10.0%
  two  18      36.0%  90.0%

But with real data I think casting to int is not recommended, better is use round.

Extra Credit:

table['% of Total'] = (table.C / table.C.sum() * 100)
table['% of B'] = (table.C / table.groupby(level=0).C.transform(sum) * 100)
table.loc['total', :] = table.sum().values
print table
              C  % of Total  % of B
A     B                            
x     one   2.0         4.0    20.0
      two   8.0        16.0    80.0
y     one  18.0        36.0    90.0
      two   2.0         4.0    10.0
z     one   2.0         4.0    10.0
      two  18.0        36.0    90.0
total      50.0       100.0   300.0

Answer 2

If you want to chain the methods to assign the new columns to pivot_table() method to put in a pipeline, you can do so using assign().

Moreover, you can add the totals as a new row using the margins parameter of pivot_table.

table = (
    df
    .pivot_table(index=['A', 'B'], aggfunc=np.sum, margins=True, margins_name='Total')
    .assign(**{
        # must exclude the last row (which are the Totals) for sum and group-specific sum
        '% of Total': lambda x: x['C'] / x.iloc[:-1]['C'].sum() * 100,
        '% of B': lambda x: x['C'] / x.iloc[:-1].groupby(level='A')['C'].transform('sum') * 100
    })
)

Note that for the particular example in the OP, as pivot_table method's columns parameter is not used, pivot_table is equivalent to groupby as explained here. So an equivalent (and possibly faster) approach to produce the initial pivot table result is

table = df.groupby(['A','B']).sum()