Fast calculation of mean with leaving one element out

Question

Lets assume I have a dataframe

df = pd.DataFrame(data={'group_id': [1, 1, 1, 1, 2, 2, 2, 2], 
    'A': [24.0, 12.0, 23.0, 22.0, 44.0, 55.0, 52.0, 48.0],
    'B': [23.0, 15.0, 22.0, 21.0, 65.0, 53.0, 53.0, 54.0]})

for every index in the dataframe I would like to calculate the mean of the group (as specified by group id) without this index included.

I started with two for loops and improved speed by using apply:

def func(x):
    df = x.copy()
    for row in x.itertuples():
        df.loc[row[0], :] = x.loc[x.index != row[0], :].mean()

    return df

df.groupby('group_id')['A', 'B'].apply(func)

The desired output is

                    A          B
group_id                        
1        0  19.000000  19.333333
         1  23.000000  22.000000
         2  19.333333  19.666667
         3  19.666667  20.000000
2        4  51.666667  53.333333
         5  48.000000  57.333333
         6  49.000000  57.333333
         7  50.333333  57.000000

Is there a faster way to compute this?

Answer 1

Use transform. Get sum and count

g = df.groupby('group_id')
sums = g.transform('sum')
counts = g.transform('count')

df[['A', 'B']].mul(-1).add(sums).div(counts - 1)

           A          B
0  19.000000  19.333333
1  23.000000  22.000000
2  19.333333  19.666667
3  19.666667  20.000000
4  51.666667  53.333333
5  48.000000  57.333333
6  49.000000  57.333333
7  50.333333  57.000000