过滤每组具有 n 个最大值的行

Question

上下文

对于每支球队，我想要包含前三名得分球员的数据框行。

在我的脑海中，它是Dataframe.nlargest() 和Dataframe.groupby() 的组合，但我不认为这是受支持的。 我理想的解决方案是：

• 直接在df 上执行，无需创建其他数据帧
• 清晰，并且
• 相对性能好（真正的 df 形状是 7M 行和 5 列）

输入

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     |

所需的输出

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 1

您可以将df.groupby 与df.nlargest 一起使用

df.groupby('team').apply(lambda x:x.nlargest(3,'points')).reset_index(drop=True)

  team   player  points
0    A    Alice      15
1    A   Carmen      13
2    A    Becky      11
3    B    Greta      29
4    B     Fran      28
5    B     Iris      25
6    C     Lucy      23
7    C    Molly      18
8    C  Ophelia      15

Answer 2

您可以使用df.groupby.rank方法：

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

Answer 3

这样的事情可能会奏效 -

df.loc[df.groupby(['team'])['points'].nlargest(3).reset_index().drop(['team','points'], axis=1)['level_1'].values]
   team   player  points
0     A    Alice      15
2     A   Carmen      13
1     A    Becky      11
6     B    Greta      29
5     B     Fran      28
8     B     Iris      25
11    C     Lucy      23
12    C    Molly      18
14    C  Ophelia      15

Answer 4

另一种方式是sort_values和groupby().tail/head：

df.sort_values('points').groupby('team').tail(3)

输出：

   team   player  points
1     A    Becky      11
2     A   Carmen      13
0     A    Alice      15
14    C  Ophelia      15
12    C    Molly      18
11    C     Lucy      23
8     B     Iris      25
5     B     Fran      28
6     B    Greta      29

或者

df.sort_values('points', ascending=False).groupby('team').head(3)

输出：

   team   player  points
6     B    Greta      29
5     B     Fran      28
8     B     Iris      25
11    C     Lucy      23
12    C    Molly      18
0     A    Alice      15
14    C  Ophelia      15
2     A   Carmen      13
1     A    Becky      11