Group By datatype screwed when group consists of one value

[gerigk]

x = DataFrame(np.arange(9).reshape(3,3))
x['test']=0
x['fl']= [1.3,1.5,1.6]
x.groupby('test').agg({'fl':'sum',2:'size'}).dtypes
Out[]:
2 int64
fl object
(version 0.8)

Output in 0.7.3
Out[]:
2 int64
fl float64

The output for x['test'] = [1,2,3] works fine

[gerigk]

Apparently fl is not only an object, but also an array!

from pandas import *
import numpy as np
x = DataFrame(np.arange(9).reshape(3,3))
x['test']=0
x['fl']= [1.3,1.5,1.6]
x.groupby('test').agg({'fl':'sum',2:'size'}).dtypes
grouped = x.groupby('test').agg({'fl':'sum',2:'size'})
type(grouped['fl'][0])

output: type 'numpy.ndarray'

print type(grouped[2][0])

output: type 'numpy.int64'

This issue is bugging me a lot because I am doing aggregations on data which may or may not have more than one category per group.

print type(grouped['fl'][0].astype(float))

still returns type 'numpy.ndarray'

whereas it works for the int column:

print type(grouped[2][0].astype(float))

output: type 'numpy.float64'

[gerigk]

if grouped.shape[0]==1:
grouped = DataFrame([tuple([grouped[y].iget(0).item() if isinstance(grouped[y].iget(0),np.ndarray) else grouped[y].iget(0) for y in grouped.columns])], index=grouped.index, columns=grouped.columns)
is an ugly workaround in case somebody else walks into this

[wesm]

Thanks for all the info, got this sorted