Add nunique and unique to series apply whitelist

[bburan-galenea]

The unique and nunique attributes are very useful in conjunction with series groupby operations. I used these extensively in previous versions of Pandas whenever I needed to get a list of unique values for each subgroup (or the number of unique values). This can be used, for example, to count the number of subjects in each treatment group (or get a list of the subject IDs for reporting):

data = pandas.DataFrame({
    'subject_id': ('A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J'),
    'treatment': (0, 0, 0, 0, 0, 1, 1, 1, 0, 0),
})
print data.groupby('treatment').subject_id.apply(lambda x: x.nunique())
print data.groupby('treatment').subject_id.apply(lambda x: x.unique())

We'd get the following output:

treatment
0            7
1            3
dtype: int64

treatment
0            [A, B, C, D, E, I, J]
1                        [F, G, H]
dtype: object

This is super-useful for generating summary statistics (e.g. N's) and debugging (e.g. tracking down which subjects are in which groups. In previous versions of Pandas, we could simply do:

print data.groupby('treatment').subject_id.nunique()
print data.groupby('treatment').subject_id.unique()

It would be nice to continue this. Is there a reason why nunique and unique can't be added to the whitelist?

[jreback]

no....care to do a quick PR (just update the whitelist and the tests for it)...

FYI

you can always do:

data.groupby(...).apply(lambda x: x.unique())

and that is excatly what is happenin in this case (its implemented essentially as a generated function)