Pandas: grouping and aggregation with multiple functions

Question

Situation

I have a pandas dataframe defined as follows:

import pandas as pd

headers = [\'Group\', \'Element\', \'Case\', \'Score\', \'Evaluation\'         


        

Answer 1

You can use apply instead of agg to construct all the columns in one go.

result = (
    df.groupby('Group').apply(lambda x: [np.max(x.Score),
                              df.loc[x.Score.idxmax(),'Element'],
                              df.loc[x.Score.idxmax(),'Case'],
                              np.min(x.Evaluation)])
      .apply(pd.Series)
      .rename(columns={0:'Max_score_value',
                       1:'Max_score_element',
                       2:'Max_score_case',
                       3:'Min_evaluation'})
      .reset_index()
)



result
Out[9]: 
  Group  Max_score_value  Max_score_element Max_score_case  Min_evaluation
0     A             9.19                  1              y            0.41
1     B             9.12                  2              x            0.10

Answer 2

Here is possible solution with pd.merge

>> r = df.groupby('Group') \
>>       .agg({'Score': 'idxmax', 'Evaluation': 'min'}) \
>>       .rename(columns={'Score': 'idx'})
>> for c in ['Score', 'Element', 'Case']:
>>   r = pd.merge(r, df[[c]], how='left', left_on='idx', right_index=True)
>> r.drop('Score_idx', axis=1).rename(columns={'Score': 'Max_score_value', 
>>                                             'Element': 'Max_score_element', 
>>                                             'Case': 'Max_score_case'})
       Evaluation  Max_score_value  Max_score_element Max_score_case
Group                                                               
A            0.41             9.19                  1              y
B            0.10             9.12                  2              x

Though it provides the desired output, I am not sure about if it's not less efficient than yours approach.

Answer 3

Starting from the result data frame, you can transform in two steps as follows to the format you need:

# collapse multi index column to single level column
result.columns = [y + '_' + x if y != '' else x for x, y in result.columns]
​
# split the idxmax column into two columns
result = result.assign(
    max_score_element = result.idxmax_Score.str[0],
    max_score_case = result.idxmax_Score.str[1]
).drop('idxmax_Score', 1)

result

#Group  max_Score   min_Evaluation  max_score_case  max_score_element
#0   A       9.19             0.41               y                  1
#1   B       9.12             0.10               x                  2

---

An alternative starting from original df using join, which may not be as efficient but less verbose similar to @tarashypka's idea:

(df.groupby('Group')
   .agg({'Score': 'idxmax', 'Evaluation': 'min'})
   .set_index('Score')
   .join(df.drop('Evaluation',1))
   .reset_index(drop=True))

#Evaluation  Group  Element   Case  Score
#0     0.41      A        1      y   9.19
#1     0.10      B        2      x   9.12

---

Naive timing with the example data set:

%%timeit 
(df.groupby('Group')
 .agg({'Score': 'idxmax', 'Evaluation': 'min'})
 .set_index('Score')
 .join(df.drop('Evaluation',1))
 .reset_index(drop=True))
# 100 loops, best of 3: 3.47 ms per loop

%%timeit
result = (
    df.set_index(['Element', 'Case'])
    .groupby('Group')
    .agg({'Score': ['max', 'idxmax'], 'Evaluation': 'min'})
    .reset_index()
)
​
result.columns = [y + '_' + x if y != '' else x for x, y in result.columns]
​
result = result.assign(
    max_score_element = result.idxmax_Score.str[0],
    max_score_case = result.idxmax_Score.str[1]
).drop('idxmax_Score', 1)
# 100 loops, best of 3: 7.61 ms per loop

Answer 4

My Take

g = df.set_index('Group').groupby(level='Group', group_keys=False)

result = g.apply(
    pd.DataFrame.nlargest, n=1, columns='Score'
)

def f(x):
    x = 'value' if x == 'Score' else x
    return 'Max_score_' + x.lower()

result.drop('Evaluation', 1).rename(columns=f).assign(
    Min_evaluation=g.Evaluation.min().values).reset_index()

  Group  Max_score_element Max_score_case  Max_score_value  Min_evaluation
0     A                  1              y             9.19            0.41
1     B                  2              x             9.12            0.10