How can I compute the absolute sum with a groupby in pandas?

Question

How can I compute the absolute sum with a groupby in pandas?

For example, given the DataFrame:

    Player  Score
0      A    100
1      B   -150
2      A   -110
3      B    180
4      B    125

I would like to have the total score for player A (100+110=210) as well as the total score for player A (150+180+125=455), ignoring the sign of the score.

I can use the following code to compute the sum:

import pandas as pd
import numpy as np

frame = pd.DataFrame({'Player' : ['A', 'B', 'A', 'B', 'B'], 
                      'Score'  : [100, -150, -110, 180, 125]})

print('frame: {0}'.format(frame))

total_scores = frame[['Player','Score']].groupby(['Player']).agg(['sum'])

print('total_scores: {0}'.format(total_scores))

but how can I compute the absolute sum with a groupby?

frame[['Player','Score']].abs().groupby(['Player']).agg(['sum']) unsurprisingly returns:

Traceback (most recent call last):
  File "O:\tests\absolute_count.py", line 10, in <module>
    total_scores = frame[['Player','Score']].abs().groupby(['Player']).agg(['sum'])
  File "C:\Users\dernoncourt\AppData\Local\Continuum\Anaconda3\lib\site-packages\pandas\core\generic.py", line 5518, in abs
    return np.abs(self)
TypeError: bad operand type for abs(): 'str'

I don't want to alter the DataFrame.

Answer 1

You could apply a function that takes the absolute value and then sums it:

>>> frame.groupby('Player').Score.apply(lambda c: c.abs().sum())
Player
A    210
B    455
Name: Score, dtype: int64

You could also create a new column with the absolute values and then sum that:

>>> frame.assign(AbsScore=frame.Score.abs()).groupby('Player').AbsScore.sum()
Player
A    210
B    455
Name: AbsScore, dtype: int64

Answer 2

You can use DataFrameGroupBy.apply with a lambda:

In [326]: df.groupby('Player').Score.apply(lambda x: np.sum(np.abs(x)))
Out[326]: 
Player
A    210
B    455
Name: Score, dtype: int64

To get back the Player column, use df.reset_index:

In [371]: df.groupby('Player').Score.apply(lambda x: np.sum(np.abs(x))).reset_index()
Out[371]: 
  Player  Score
0      A    210
1      B    455