Panda dataframe conditional .mean() depending on values in certain column

Question

I'm trying to create a new column which returns the mean of values from an existing column in the same df. However the mean should be computed based on a grouping in three other columns.

Out[184]: 
   YEAR daytype hourtype  scenario  option_value    
0  2015     SAT     of_h         0      0.134499       
1  2015     SUN     of_h         1     63.019250      
2  2015     WD      of_h         2     52.113516       
3  2015     WD      pk_h         3     43.126513       
4  2015     SAT     of_h         4     56.431392 

I basically would like to have a new column 'mean' which compute the mean of "option value", when "YEAR", "daytype", and "hourtype" are similar.

I tried the following approach but without success ...

In [185]: o2['premium']=o2.groupby(['YEAR', 'daytype', 'hourtype'])['option_cf'].mean()

TypeError: incompatible index of inserted column with frame index

Answer 1

Here's one way to do it

In [19]: def cust_mean(grp):
   ....:     grp['mean'] = grp['option_value'].mean()
   ....:     return grp
   ....:

In [20]: o2.groupby(['YEAR', 'daytype', 'hourtype']).apply(cust_mean)
Out[20]:
   YEAR daytype hourtype  scenario  option_value       mean
0  2015     SAT     of_h         0      0.134499  28.282946
1  2015     SUN     of_h         1     63.019250  63.019250
2  2015      WD     of_h         2     52.113516  52.113516
3  2015      WD     pk_h         3     43.126513  43.126513
4  2015     SAT     of_h         4     56.431392  28.282946

So, what was going wrong with your attempt?

It returns an aggregate with different shape from the original dataframe.

In [21]: o2.groupby(['YEAR', 'daytype', 'hourtype'])['option_value'].mean()
Out[21]:
YEAR  daytype  hourtype
2015  SAT      of_h        28.282946
      SUN      of_h        63.019250
      WD       of_h        52.113516
               pk_h        43.126513
Name: option_value, dtype: float64

Or use transform

In [1461]: o2['premium'] = (o2.groupby(['YEAR', 'daytype', 'hourtype'])['option_value']
                              .transform('mean'))

In [1462]: o2
Out[1462]:
   YEAR daytype hourtype  scenario  option_value    premium
0  2015     SAT     of_h         0      0.134499  28.282946
1  2015     SUN     of_h         1     63.019250  63.019250
2  2015      WD     of_h         2     52.113516  52.113516
3  2015      WD     pk_h         3     43.126513  43.126513
4  2015     SAT     of_h         4     56.431392  28.282946

Answer 2

You can do it the way you intended by tweaking your code in the following way:

o2 = o2.set_index(['YEAR', 'daytype', 'hourtype'])

o2['premium'] = o2.groupby(level=['YEAR', 'daytype', 'hourtype'])['option_value'].mean()

Why the original error? As explained by John Galt, the data coming out of groupby().mean() is not the same shape (length) as the original DataFrame.

Pandas can handle this cleverly if you first start with the 'grouping columns' in the index. Then it knows how to propogate the mean data correctly.

John's solution follows the same logic, because groupby naturally puts the grouping columns in the index during execution.