Pandas custom second level groupby function-CodePudding

I have this:

df = pd.DataFrame({'sku_id' : ['A','A','A','B','C','C'],
                  'order_counts' : [1,2,3,1,1,2],
                  'order_val' : [10,20,30,10,10,20]})

which creates:

A simple groupby() of sku_id using df.groupby('sku_id').sum() would give:

And a two-level grouping df.groupby(['sku_id', 'order_counts']).sum() would give:

But now I want a custom second-level grouping on order_counts such that any order_counts == 1 is considered a group and any order_counts > 1 are grouped together in a group labelled R (for Repeat order)

The result would look like this:

sku_id order_counts  order_val         
A      1                    10
       R                    50                           
B      1                    10
C      1                    10
       R                    20

Is there a way to supply a custom groupby function to achieve this?

CodePudding user response：

Mask the != 1 values in the order_counts column with R, then use groupby sum

g = df['order_counts'].mask(df['order_counts'] != 1, 'R')
df.groupby(['sku_id', g])['order_val'].sum()

Result

sku_id  order_counts
A       1               10
        R               50
B       1               10
C       1               10
        R               20
Name: order_val, dtype: int64

CodePudding user response：

How about just assign before you groupby?

new_df = (df
    .assign(order_counts=lambda x: np.where(x['order_counts'] > 1, 'R', x['order_counts']))
    .groupby(['sku_id', 'order_counts'])
    .sum()
)

Output:

>>> new_df
                     order_val
sku_id order_counts           
A      1                    10
       R                    50
B      1                    10
C      1                    10
       R                    20