Inconsistent behavior of apply with operator.itemgetter v.s. applymap operator.itemgetter

Question

May not be the best way of arrange data in real case, but it serves a good example:

In [16]:
import operator
In [17]:
DF=pd.DataFrame({'Val1':[[2013, 37722.322],[1998, 32323.232]],
                 'Val2':[[2013, 37722.322],[1998, 32323.232]]})
In [18]:
print DF
                Val1               Val2
0  [2013, 37722.322]  [2013, 37722.322]
1  [1998, 32323.232]  [1998, 32323.232]

[2 rows x 2 columns]

apply gives wrong result

In [19]:
print DF.apply(operator.itemgetter(-1), axis=1)
   Val1       Val2
0  2013  37722.322
1  1998  32323.232

[2 rows x 2 columns]

but applymap gives the right result!

In [20]:
print DF.applymap(operator.itemgetter(-1))
        Val1       Val2
0  37722.322  37722.322
1  32323.232  32323.232

[2 rows x 2 columns]

Why could it happen?

Answer 1

It is easier to see what is happening if you use

df = pd.DataFrame({'Val1':[[1, 2],[3, 4]],
                 'Val2':[[5, 6],[7, 8]]})

     Val1    Val2
0  [1, 2]  [5, 6]
1  [3, 4]  [7, 8]

df.apply(operator.itemgetter(-1), axis=1) calls operator.itemgetter(-1) on each row.

For example, on the first row, operator.itemgetter(-1) returns the last item, which is [5, 6]. Since this value is iterable, its values are then assigned to the two columns Val1 and Val2. So the result is

In [149]: df.apply(operator.itemgetter(-1), axis=1)
Out[149]: 
   Val1  Val2
0     5     6
1     7     8

In contrast, applymap operates on each cell in the DataFrame individually, so operator.itemgetter(-1) returns the last item from each cell.

In [150]: df.applymap(operator.itemgetter(-1))
Out[150]: 
   Val1  Val2
0     2     6
1     4     8