BUG: Add squeeze keyword to groupby to allow reduction in returned type

[jreback]

This will allow a reduction in the returned type from DataFrame -> Series
if the groups are unique.

This is a fix for a regression from 0.10.1.

Allows functionaility in #2893, by specifying squeeze=True
in groupby call. #3596 functionaility is back as the default

This returns a Series because we are passing squeeze=True

In [9]: df2 = DataFrame([{"val1": 1, "val2" : 20}, 
{"val1":1, "val2": 19},{"val1":1, "val2": 27}, {"val1":1, "val2": 12}])

In [10]: df2
Out[10]: 
   val1  val2
0     1    20
1     1    19
2     1    27
3     1    12

In [11]: def func(dataf):
   ....:     return dataf["val2"]  - dataf["val2"].mean()
   ....: 

In [12]: df2.groupby("val1", squeeze=True).apply(func)
Out[12]: 
0    0.5
1   -0.5
2    7.5
3   -7.5
Name: 1, dtype: float64

Traditionally returns a DataFrame (even though have unique groups)
Implicity (squeeze=False)

In [13]: df = DataFrame([[1,1],[1,1]],columns=['X','Y'])

In [14]: df
Out[14]: 
   X  Y
0  1  1
1  1  1

In [15]: df.groupby('X').count()
Out[15]: 
   X  Y
X      
1  2  2

[jreback]

@wesm pls take a look. I think the prior behavior is correct, most often you would not want the reduction (which is data dependent). The new keyword allows it.

[hayd]

This is a good feature/fix.
Note: squeeze is what's used in read_csv to do this... perhaps we should use the same name here?

[jreback]

thats a good idea! will update. thankxs

[jreback]

@hayd better? (and is the example in v0.11.1 adequate?)