filter for rows with n largest values for each group

Context

I want, for each team, the rows of the data frame that contains the top three scoring players.

In my head, it is a combination of Dataframe.nlargest() and Dataframe.groupby() but I don’t think this is supported. My ideal solution is:

• performed directly on df without having to create other dataframes
• legible, and
• relatively performant (real df shape is 7M rows and 5 col)

Input

import pandas as pd
df = pd.read_json('{"team":{"0":"A","1":"A","2":"A","3":"A","4":"A","5":"B","6":"B","7":"B","8":"B","9":"B","10":"C","11":"C","12":"C","13":"C","14":"C"},"player":{"0":"Alice","1":"Becky","2":"Carmen","3":"Donna","4":"Elizabeth","5":"Fran","6":"Greta","7":"Heather","8":"Iris","9":"Jackie","10":"Kelly","11":"Lucy","12":"Molly","13":"Nina","14":"Ophelia"},"points":{"0":15,"1":11,"2":13,"3":8,"4":10,"5":28,"6":29,"7":18,"8":25,"9":9,"10":12,"11":23,"12":18,"13":10,"14":15}}')

| team | player    | points |
|------|-----------|--------|
| A    | Alice     | 15     |
| A    | Becky     | 11     |
| A    | Carmen    | 13     |
| A    | Donna     | 8      |
| A    | Elizabeth | 10     |
| B    | Fran      | 28     |
| B    | Greta     | 29     |
| B    | Heather   | 18     |
| B    | Iris      | 25     |
| B    | Jackie    | 9      |
| C    | Kelly     | 12     |
| C    | Lucy      | 23     |
| C    | Molly     | 18     |
| C    | Nina      | 10     |
| C    | Ophelia   | 15     |

Desired Output

df_output = pd.read_json('{"team":{"0":"A","1":"A","2":"A","3":"B","4":"B","5":"B","6":"C","7":"C","8":"C"},"player":{"0":"Alice","1":"Becky","2":"Carmen","3":"Fran","4":"Greta","5":"Iris","6":"Lucy","7":"Molly","8":"Ophelia"},"points":{"0":15,"1":11,"2":13,"3":28,"4":29,"5":25,"6":23,"7":18,"8":15}}')
df_output

| team | player  | points |
|------|---------|--------|
| A    | Alice   | 15     |
| A    | Becky   | 11     |
| A    | Carmen  | 13     |
| B    | Fran    | 28     |
| B    | Greta   | 29     |
| B    | Iris    | 25     |
| C    | Lucy    | 23     |
| C    | Molly   | 18     |
| C    | Ophelia | 15     |

Answer

You can use df.groupby.rank method:

In [1401]: df[df.groupby('team')['points'].rank(ascending=False) <= 3]
Out[1401]: 
   team   player  points
0     A    Alice      15
1     A    Becky      11
2     A   Carmen      13
5     B     Fran      28
6     B    Greta      29
8     B     Iris      25
11    C     Lucy      23
12    C    Molly      18
14    C  Ophelia      15