Groupby sum and difference of rows in a pandas dataframe - pandas

Question

I have a dataframe:
df = pd.DataFrame({
'Metric': ['Total Assets', 'Total Promo', 'Total Assets', 'Total Int'],
'Product': ['AA', 'AA', 'BB', 'AA'],
'Risk': ['High', 'High','Low', 'High'],
'202101': [ 130, 200, 190, 210],
'202102': [ 130, 200, 190, 210],
'202103': [ 130, 200, 190, 210],})
I would like to groupby Product and Risk and sum the entries in Total Assets and Total Promo and subtract the result from the entries in Total Int. I could multiply all rows with Total Int by -1 and sum the result. But I wanted to know if there was a direct way to do so.
df.groupby(['Product', 'Risk']).sum()
The actual dataset is large and it would introduce complexity to multiply certain rows by -1
The output would look like:
df = pd.DataFrame({
'Product': ['AA', 'BB'],
'Risk': ['High', 'Low'],
'202101': [ 120, 190],
'202102': [ 120, 190],
'202103': [ 120, 190],})

Answer 1

You can multiply by -1 your Total Int rows:
df.loc[df['Metric'] == 'Total Int', df.select_dtypes('number').columns] *= -1
# OR
df.loc[df['Metric'] == 'Total Int', df.filter(regex=r'\d{6}').columns] *= -1
>>> df.groupby(['Product', 'Risk']).sum()
202101 202102 202103
Product Risk
AA High 120 120 120
BB Low 190 190 190

Answer 2

In your actual dataset, do you have any groups that only have one row? The following solution will work if all groups have greater than one row, so that diff(), doesn't return nan. This is thy the second row of output is not in there, but I imagine your groups have more than one row in your large dataset.
IIUC, create a series s that differentiates the two groups and take the diff after a groupby of the sum:
import pandas as pd
import numpy as np
df = pd.DataFrame({
'Metric': ['Total Assets', 'Total Promo', 'Total Assets', 'Total Int'],
'Product': ['AA', 'AA', 'BB', 'AA'],
'Risk': ['High', 'High','Low', 'High'],
'Col1': [ 130, 200, 190, 210],
'Col2': [ 130, 200, 190, 210],
'Col3': [ 130, 200, 190, 210],})
s = np.where(df['Metric'].isin(['Total Assets', 'Total Promo']), 'B', 'A')
cols = ['Product', 'Risk']
(df.groupby(cols + [s]).sum()
.groupby(cols).diff()
.dropna().reset_index().drop('level_2', axis=1))
Out[1]:
Product Risk Col1 Col2 Col3
0 AA High 120.0 120.0 120.0

Answer 3

How about this as a solution?
(df.
melt(['Metric', 'Product', 'Risk']).
pivot(index=['Product', 'Risk', 'variable'], columns= 'Metric', values = 'value').
assign(Total = lambda df: df['Total Assets'].fillna(0)+df['Total Promo'].fillna(0) - df['Total Int'].fillna(0)).
drop(columns = ['Total Assets', 'Total Promo', 'Total Int']).
reset_index().
pivot(index=['Product', 'Risk'], columns= 'variable', values = 'Total')
)