ENH: Create a nlargest and nsmallest methods for pandas.core.groupby.DataFrameGroupBy objects

[mcarbajo]

Code Sample, a copy-pastable example if possible

Current method

import pandas as pd

dataframe = pd.DataFrame({'col1': list('aaabbbcc'), 'col2': range(8), 'col3': list('whatever')})
dataframe.groupby('col1').apply(lambda df: df.nlargest(2, 'col2'))

Desired method

import pandas as pd

dataframe = pd.DataFrame({'col1': list('aaabbbcc'), 'col2': range(8), 'col3': list('whatever')})
dataframe.groupby('col1').nlargest(2, 'col2')

Problem description

I want to select the first 2 entries of each group ordered by a specific column. I can easily get these values if I only want to keep the column used for the ordering (see code below) but it gets way less efficient and not as clear when I want to keep all the columns of the DataFrame.

import pandas as pd

dataframe = pd.DataFrame({'col1': list('aaabbbcc'), 'col2': range(8), 'col3': list('whatever')})
dataframe.groupby('col1').col2.nlargest(2)

[WillAyd]

Might be overlooking it but what is wrong with using apply? We have very few groupby functions outside of apply that either don't maintain the shape of the caller or reduce to a scalar per group, so this would be somewhat of an anomaly. As such -1 as I don't see the value add against using apply

[mroeschke]

I'm also -1 for the same reason as Will. Since this issue hasn't gotten much traction, closing.