Pandas pivot_table percentile

Question

I'm trying to describe columns A, B by mean, median, 25th percentile, 75th percentile, standard deviation.

df = pd.DataFrame({'A':[1,9,3,4,6,8,2,7],
                   'B':[2,4,7,8,9,2,5,6],
                   'S':['L','L','L','S','L','S','S','L']})

Here is what I did and it worked since I only have 25th percentile:

df.pivot_table(columns = ['S'], values = ['A','B'], aggfunc = [np.mean, lambda x: np.percentile(x,25), np.median, np.std])

But if I also put the 75th percentile in, it gives me the error message:

Reindexing only valid with uniquely valued Index objects

Ideally I would like the output list the 75th percentile in the next columns.

Answer 1

This will do what I think you want, but without a lambda and few extra lines:

def my25(g):
    return np.percentile(g, 25)

def my75(g):
    return np.percentile(g, 75)

df.pivot_table(columns = ['S'], values = ['A','B'], 
               aggfunc = [np.mean, my25, np.median, np.std, my75])

        mean      my25    median         std      my75     
S    L     S    L    S      L  S     L     S    L    S
A  5.2  4.67    3  3.0      6  4  3.19  3.06    7  6.0
B  5.6  5.00    4  3.5      6  5  2.70  3.00    7  6.5

EDIT: actually, it is possible to use only lambda functions if you use groupby to aggregate instead of pivot_table, and supply a name to each function.

func_lst = [('mean',np.mean), ('25',lambda x:np.percentile(x,0.25)), 
            ('med',np.median), ('std',np.std), ('75',lambda x:np.percentile(x,0.75))]

df.groupby('S').agg(func_lst).stack(level=0).unstack(level=0).swaplevel(0,1,axis=1)

        mean      25    med         std      75     
S    L     S  L    S   L  S     L     S  L    S
A  5.2  4.67  3  3.0   6  4  3.19  3.06  7  6.0
B  5.6  5.00  4  3.5   6  5  2.70  3.00  7  6.5

I thought using func_lst in a pivot_table call might work but it doesn't. Anyway to me it is clearer to just define the my25, my75 functions and use the pivot_table.