Change groupby value_counts (from fall through behaviour)

[hayd]

The fall through value_counts (for Series) is a bit strange, I think better result would be (with the standard options):

Can put together if people think it's good.

In [132]: df = pd.DataFrame([['a_link', 'dofollow'], ['a_link', 'dofollow'], ['a_link', 'nofollow'], ['b_link', 'javascript']], columns=['link', 'type'])

In [133]: g = df.groupby(['link', 'type'])

In [134]: g.value_counts()
AttributeError: 'DataFrameGroupBy' object has no attribute 'value_counts'

In [135]: g.link.value_counts()   # redundant level
Out[135]: 
link    type              
a_link  dofollow    a_link    2
        nofollow    a_link    1
b_link  javascript  b_link    1
dtype: int64


Following would make sense for DataFrameGroupby to:
In [136]: pd.Series([len(g.groups[i]) for i in g.grouper.result_index], g.grouper.result_index)
Out[136]: 
link    type      
a_link  dofollow      2
        nofollow      1
b_link  javascript    1
dtype: int64

Note: as_index doesn't make sense here so would be ignored.

[hayd]

Ahem, that would be size....

[hayd]

Maybe it should be size with standard value_count options ?

[jreback]

maybe just alias value_counts to size?

this related (maybe dupe) of #6312

[hayd]

Yeah I think this makes sense, I don't think most people want to do anything else (trying to think of a case when this wouldn't be desired)... I guess that's when the selected column isn't in the index... darn then it is going to be different. :s