CODEWITHSUNDEEP
pythonpandas
pasternak
pasternak

Reputation: 403

Summarize using multiple columns in python pandas dataframe

In R, I can summarize the data using more than one data columns as follows: library(dplyr):

A =  B %>%
group_by(col1,col2) %>%
summarize(newcol = sum(col3)/sum(col4))

But in python's pandas' dataframes, how do I perform the same operation in one step?

I can do this in two steps. Step 1:

A = B.groupby(['col1','col2']).agg({'col3': 'sum','col4':'sum'})

Step 2:

A['newcol'] = A['col3']/A['col4']

Upvotes: 4

Views: 3687

Answers (2)

Panwen Wang
Panwen Wang

Reputation: 3825

With datar, you can do it the same way you did in R:

from datar import f
from datar.base import sum
from datar.dplyr import group_by, summarise

A =  (
  B 
  >> group_by(f.col1,f.col2) 
  >> summarize(newcol = sum(f.col3)/sum(f.col4))
)

Upvotes: 1

user2285236
user2285236

Reputation:

You need to use assign with a lambda expression:

df = pd.DataFrame({'col1': list('aaabbb'), 
                   'col2': list('xyxyxy'), 
                   'col3': np.random.randn(6), 
                   'col4': np.random.randn(6)})

df
Out: 
  col1 col2      col3      col4
0    a    x -2.276155  0.323778
1    a    y -0.367525 -2.570142
2    a    x -0.672530  2.265560
3    b    y  0.588741  0.193499
4    b    x -1.368829  0.717997
5    b    y  1.012271  1.354408

(df.groupby(['col1','col2'])
   .agg({'col3': 'sum','col4':'sum'})
   .assign(newcol=lambda x: x['col3']/x['col4']))
Out: 
               col4      col3    newcol
col1 col2                              
a    x     2.589338 -2.948686 -1.138780
     y    -2.570142 -0.367525  0.142998
b    x     0.717997 -1.368829 -1.906453
     y     1.547907  1.601012  1.034308

If all you need is the new column, use apply:

df.groupby(['col1','col2']).apply(lambda x: x['col3'].sum() / x['col4'].sum())
Out: 
col1  col2
a     x      -1.138780
      y       0.142998
b     x      -1.906453
      y       1.034308
dtype: float64

If you are using this on a big data set, avoid apply and use eval instead.

(df.groupby(['col1','col2'])
   .agg({'col3': 'sum','col4':'sum'})
   .eval('col3 / col4'))

Upvotes: 7

Related Questions