Add aggregate method to DataFrame

[wesm]

a la groupby-aggregate

[gerigk]

I don't know whether this is a separate issue but a similar request would be to be able to use a non-unique 1d index for groupby.
If I have a MultiIndex I can do df.groupby(level = 'A'), but if the index is only 'A' I have to drop the index and then group by the column.
My concrete use case in this context is df.groupby(level='A').first() where 'A' is the index. It fails whenever I happen to have a 1d index.

[jreback]

should be easy after #11841

[nbonnotte]

This may be a silly question, but I'm gonna ask it: what should be the result of this DataFrame.aggregate?

For instance, let's consider df.aggregate(np.sum).
Should the result be like df.sum()?
Or like df.groupby(df.index).sum()?
Or something else?

[jreback]

this would work like .groupby(...).agg(np.sum) in the sense that the aggregations will take the same form. But this is an aggregation on the frame itself.

e.g. imagine

df.agg(['sum', 'count'])
df.agg({'A' : 'sum', 'B' : 'count'})
df.agg({'A' : ['sum','count'], 'B' : ['sum','mean']})

and equivalent for:

df.A.agg(['sum','count'])

and so on

[nbonnotte]

So there would be no grouping at all?

In a sense, it would be a generalisation of DataFrame.apply, except that it would work also with strings and be able to yield more complex output?

[jreback]

yes .apply and .aggregate are synonymous for .groupby,.rolling, .resample already, so having the same for NDFrame is a good idea (btw this is quite easy to implement as all of the machinery is already there).