使用groupby pandas计算行数

Question

I had the following function in pandas 0.17:

df['numberrows'] = df.groupby(['column1','column2','column3'], as_index=False)[['column1']].transform('count').astype('int')

But I upgraded pandas today and now I get the error:

  File "/usr/local/lib/python3.4/dist-packages/pandas/core/internals.py", 

line 3810, in insert raise ValueError('cannot insert {}, already exists'.format(item))

ValueError: cannot insert column1, already exists

What has changed in the update which causes this function to not work anymore?

I want to groupby the columns and add a column which has the amount or rows of the groupby.

If what I did before was not a good function, another way of grouping while getting the amount of rows that were grouped is also welcome.

EDIT:

small dataset:

    column1  column2    column3   
 0  test     car1       1           
 1  test2    car5       2         
 2  test     car1       1         
 3  test4    car2       1      
 4  test2    car1       1         

outcome would be:

    column1  column2    column3   numberrows
 0  test     car1       1           2
 1  test2    car5       2           1     
 3  test4    car2       1           1
 4  test2    car1       1           1

Answer 1

Consider the following approach:

In [18]: df['new'] = df.groupby(['column1','column2','column3'])['column1'] \
                       .transform('count')

In [19]: df
Out[19]:
  column1 column2  column3  new
0    test    car1        1    2
1   test2    car5        2    1
2    test    car1        1    2
3   test4    car2        1    1
4   test2    car1        1    1

UPDATE:

In [26]: df.groupby(['column1','column2','column3'])['column1'] \
           .count().reset_index(name='numberrows')
Out[26]:
  column1 column2  column3  numberrows
0    test    car1        1           2
1   test2    car1        1           1
2   test2    car5        2           1
3   test4    car2        1           1

Answer 2

Your syntax is sloppy, you are using as_index=False with transform .
as_index=False will end up pushing those columns back into the dataframe proper when it finds that column1 already exists... uh-oh. However, that is completely unnecessary as transform handles the index for you.

df.groupby(
    ['column1','column2','column3']
)['column1'].transform('count').astype('int')

0    2
1    1
2    2
3    1
4    1
Name: column1, dtype: int64

Or make a new column

df.assign(
    new=df.groupby(
        ['column1','column2','column3']
    )['column1'].transform('count').astype('int')
)

  column1 column2  column3  new
0    test    car1        1    2
1   test2    car5        2    1
2    test    car1        1    2
3   test4    car2        1    1
4   test2    car1        1    1