Ttyjukl

How can I create new rows from an existing DataFrame by grouping by certain fields (in the example "Country" and "Industry") and applying some math to another field (in the example "Field" and "Value")?

Source DataFrame

df = pd.DataFrame({'Country': ['USA','USA','USA','USA','USA','USA','Canada','Canada'],

                   'Industry': ['Finance', 'Finance', 'Retail', 

                                'Retail', 'Energy', 'Energy', 

                                'Retail', 'Retail'],

                   'Field': ['Import', 'Export','Import', 

                             'Export','Import', 'Export',

                             'Import', 'Export'],

                   'Value': [100, 50, 80, 10, 20, 5, 30, 10]})



    Country Industry    Field   Value

0   USA     Finance     Import  100

1   USA     Finance     Export  50

2   USA     Retail      Import  80

3   USA     Retail      Export  10

4   USA     Energy      Import  20

5   USA     Energy      Export  5

6   Canada  Retail      Import  30

7   Canada  Retail      Export  10

Target DataFrame

Net = Import - Export

    Country Industry    Field   Value

0   USA     Finance     Net     50

1   USA     Retail      Net     70

2   USA     Energy      Net     15

3   Canada  Retail      Net     20

There are quite possibly many ways. Here's one using groupby and unstack:

(df.groupby(['Country', 'Industry', 'Field'], sort=False)['Value']

   .sum()

   .unstack('Field')

   .eval('Import - Export')

   .reset_index(name='Value'))



  Country Industry  Value

0     USA  Finance     50

1     USA   Retail     70

2     USA   Energy     15

3  Canada   Retail     20

IIUC

df=df.set_index(['Country','Industry'])



Newdf=(df.loc[df.Field=='Export','Value']-df.loc[df.Field=='Import','Value']).reset_index().assign(Field='Net')

Newdf

  Country Industry  Value Field

0     USA  Finance    -50   Net

1     USA   Retail    -70   Net

2     USA   Energy    -15   Net

3  Canada   Retail    -20   Net

pivot_table

df.pivot_table(index=['Country','Industry'],columns='Field',values='Value',aggfunc='sum').

  diff(axis=1).

     dropna(1).

        rename(columns={'Import':'Value'}).

          reset_index()

Out[112]: 

Field Country Industry  Value

0      Canada   Retail   20.0

1         USA   Energy   15.0

2         USA  Finance   50.0

3         USA   Retail   70.0

You can use Groupby.diff() and after that recreate the Field column and finally use DataFrame.dropna:

df['Value'] = df.groupby(['Country', 'Industry'])['Value'].diff().abs()

df['Field'] = 'Net'

df.dropna(inplace=True)

df.reset_index(drop=True, inplace=True)



print(df)

  Country Industry Field  Value

0     USA  Finance   Net   50.0

1     USA   Retail   Net   70.0

2     USA   Energy   Net   15.0

3  Canada   Retail   Net   20.0

You can do it this way to add those rows to your original dataframe:

df.set_index(['Country','Industry','Field'])

  .unstack()['Value']

  .eval('Net = Import - Export')

  .stack().rename('Value').reset_index()

Output:

   Country Industry   Field  Value

0   Canada   Retail  Export     10

1   Canada   Retail  Import     30

2   Canada   Retail     Net     20

3      USA   Energy  Export      5

4      USA   Energy  Import     20

5      USA   Energy     Net     15

6      USA  Finance  Export     50

7      USA  Finance  Import    100

8      USA  Finance     Net     50

9      USA   Retail  Export     10

10     USA   Retail  Import     80

11     USA   Retail     Net     70

This answer takes advantage of the fact that pandas puts the group keys in the multiindex of the resulting dataframe. (If there were only one group key, you could use loc.)

>>> s = df.groupby(['Country', 'Industry', 'Field'])['Value'].sum()

>>> s.xs('Import', axis=0, level='Field') - s.xs('Export', axis=0, level='Field')

Country  Industry

Canada   Retail      20

USA      Energy      15

         Finance     50

         Retail      70

Name: Value, dtype: int64