Return top N largest values per group using pandas

问题

I am trying to find maximum values in number of series each grouped by name of the column it was extracted from.

I have a dataframe as such:

MASTER      SLAVE       Value

Master_1    Slave_1     657879
Master_1    Slave_2     34343
Master_1    Slave_3     453313

Master_2    Slave_1     56667
Master_2    Slave_2     6879
Master_2    Slave_3     12333
Master_2    Slave_4     789
Master_2    Slave_5     22235

Master_3    Slave_1     65765
Master_3    Slave_2     23431
Master_3    Slave_3     445
Master_3    Slave_4     567

I need to find maximum values of first two slaves of each master.

This is so far I've gotten yet:

df.groupby('MASTER')['SLAVE'].unique()

It output series 'Slaves' values for each 'MASTER':

Master_1    [657879, 34343, 453313]
Master_2    [56667, 6879, 12333, 789, 22235]
Master_3    [65765, 23431, 445, 789, 567]

But I fail to understand what type of data I'm dealing with after this input. And how I can sort those values.

回答1:

IIUC, one option is sort_values and GroupBy.head with n=2:

df.sort_values('Value', ascending=False).groupby('MASTER', sort=False).head(2)

     MASTER    SLAVE   Value
0  Master_1  Slave_1  657879
2  Master_1  Slave_3  453313
8  Master_3  Slave_1   65765
3  Master_2  Slave_1   56667
9  Master_3  Slave_2   23431
7  Master_2  Slave_5   22235

Another is using set_index and GroupBy.nlargest with n=2:

df.set_index('SLAVE').groupby('MASTER')['Value'].nlargest(2).reset_index()

     MASTER    SLAVE   Value
0  Master_1  Slave_1  657879
1  Master_1  Slave_3  453313
2  Master_2  Slave_1   56667
3  Master_2  Slave_5   22235
4  Master_3  Slave_1   65765
5  Master_3  Slave_2   23431

回答2:

You can use a combination of sort and groupby:

df.sort_values(['MASTER', 'Value'], ascending=[True, False], inplace=True)
grp = df.groupby('MASTER')['SLAVE'].indices
slaves = {k: df.loc[k][:2]['SLAVE'].values  for k in grp.keys()}

Will output:

{'Master_1': array(['Slave_1', 'Slave_3'], dtype=object),
 'Master_2': array(['Slave_1', 'Slave_5'], dtype=object),
 'Master_3': array(['Slave_1', 'Slave_2'], dtype=object)}

来源：https://stackoverflow.com/questions/56468897/return-top-n-largest-values-per-group-using-pandas

标签

python

pandas

dataframe

group-by