Neat inter-row calculations within pandas DataFrame with MultiIndex

Question

I have the classic ucb admission dataset as a pandas DataFrame with multiIndex:

                            value
Dept      Gender  Admit     
A         Male    Admitted  512
                  Rejected  313
          Female  Admitted  89
                  Rejected  19

etc. for other departments ('A' through 'F')

and I want to create a table of the ratio of students accepted to rejected, grouped by Dept and Gender

My current approaches have been

ucbA.groupby(level=['Dept', 'Gender']).apply(lambda x: x.xs('Admitted', level=2).iloc[0] / x.xs('Rejected', level=2).iloc[0]).unstack().value

which is horrible

and

admitted = ucbA.unstack('Admit')
DataFrame({'Proportion Accepted': admitted.value.Admitted / admitted.value.Rejected}).unstack(1)

which ok I guess, but I feel it should be possible as a one-liner without unstacking.

Is there a really neat way of doing something like this? I'm imagining a one-liner staying within the context of the multi-index.

Edit: The full frame:

DataFrame({'Admit': {0: 'Admitted',  1: 'Rejected',  2: 'Admitted',  3: 'Rejected',  4: 'Admitted',  5: 'Rejected',  6: 'Admitted',  7: 'Rejected',  8: 'Admitted',  9: 'Rejected',  10: 'Admitted',  11: 'Rejected',  12: 'Admitted',  13: 'Rejected',  14: 'Admitted',  15: 'Rejected',  16: 'Admitted',  17: 'Rejected',  18: 'Admitted',  19: 'Rejected',  20: 'Admitted',  21: 'Rejected',  22: 'Admitted',  23: 'Rejected'}, 'Dept': {0: 'A',  1: 'A',  2: 'A',  3: 'A',  4: 'B',  5: 'B',  6: 'B',  7: 'B',  8: 'C',  9: 'C',  10: 'C',  11: 'C',  12: 'D',  13: 'D',  14: 'D',  15: 'D',  16: 'E',  17: 'E',  18: 'E',  19: 'E',  20: 'F',  21: 'F',  22: 'F',  23: 'F'}, 'Gender': {0: 'Male',  1: 'Male',  2: 'Female',  3: 'Female',  4: 'Male',  5: 'Male',  6: 'Female',  7: 'Female',  8: 'Male',  9: 'Male',  10: 'Female',  11: 'Female',  12: 'Male',  13: 'Male',  14: 'Female',  15: 'Female',  16: 'Male',  17: 'Male',  18: 'Female',  19: 'Female',  20: 'Male',  21: 'Male',  22: 'Female',  23: 'Female'}, 'value': {0: 512,  1: 313,  2: 89,  3: 19,  4: 353,  5: 207,  6: 17,  7: 8,  8: 120,  9: 205,  10: 202,  11: 391,  12: 138,  13: 279,  14: 131,  15: 244,  16: 53,  17: 138,  18: 94,  19: 299,  20: 22,  21: 351,  22: 24,  23: 317}}).set_index(['Dept', 'Gender', 'Admit']).astype(float).astype(int)

Alternatively if you have rpy:

import pandas.rpy.common as com
ucbA = com.load_data('UCBAdmissions').set_index(['Dept', 'Gender', 'Admit']).astype(float).astype(int)

Answer 1

Here you go:

df = pd.DataFrame({'Dept':['A','A','A','A'],
                   'Gender':['Male', 'Male', 'Female', 'Female'],
                   'Admit':['Admitted', 'Rejected', 'Admitted', 'Rejected'],
                   'value':[512,313,89,19]})
df = df.set_index(['Dept', 'Gender', 'Admit'])


# Proportions accepted and rejected:
df / df.groupby(level=['Dept','Gender']).transform(sum)
#                         value
#Dept Gender Admit             
#A    Female Admitted  0.824074
#            Rejected  0.175926
#     Male   Admitted  0.620606
#            Rejected  0.379394

# If you really want admitted as fraction of rejected:
df2 = df.swaplevel(1,2).swaplevel(0,1)
df2.ix['Admitted'] / df2.ix['Rejected']
#                value
#Dept Gender          
#A    Male    1.635783
#     Female  4.684211

Answer 2

Here's a way

In [55]: grouper = ['Dept','Gender']

In [56]: x = df.reset_index()

In [57]: (x[x.Admit=='Admitted'].groupby(grouper).sum() / 
          x[x.Admit=='Rejected'].groupby(grouper).sum()
         ).unstack()
Out[57]: 
           value          
Gender    Female      Male
Dept                      
A       4.684211  1.635783
B       2.125000  1.705314
C       0.516624  0.585366
D       0.536885  0.494624
E       0.314381  0.384058
F       0.075710  0.062678

[6 rows x 2 columns]