Question

Suppose I have a pandas DF with ‘A’,’B’,’C’ as column name

A    B   C 
a1  b11 c11
a1  b12 c12
a2  b21 c21
a2  b22 c22

I can group by ‘A’, but can I get

A  B  C
a1 [b11,b12], [c11,c12]
a2 [b21,b22], [c21,c22]

without any aggregation? Hopefully the order (b11 before b12) is kept as occured in the original table.

Answer 1

I don’t know how to do exactly what you want, but perhaps this is close enough:

In [23]: df = pd.DataFrame({'A' : ['a1', 'a1', 'a2', 'a2'],
                            'B' : ['b11', 'b12', 'b21', 'b22'],
                            'C' : ['c11', 'c12', 'c21', 'c22']})

In [24]: grpA  = df.groupby('A')
In [25]: a1 = grpA.get_group('a1')

Using that I then get:

In [26]: a1['B']  # or a1.B
Out[26]: 
0    b11
1    b12
Name: B

also:

In [39]: import numpy as np

In [40]: np.array(a1.B)
Out[40]: array([b11, b12], dtype=object)

and finally:

In [41]: grpdA.get_group('a1').B.tolist()  # leave off `.tolist()` to get a series
Out[41]: ['b11', 'b12']

Hope that helps.