如何在pandas中同时进行前后滚动？

Question

我需要检查当前值是否高于下两个和前三个值。否则，它传递前一个数字。

例如：

date	pos
2007-01-01	3.0
2007-01-02	5.0
2007-01-03	6.0
2007-01-10	11.0
2007-01-11	9.0
2007-01-20	8.0
2007-01-21	10.0
2007-01-22	13.0
2007-01-23	4.0
2007-01-27	2.0
2007-01-28	1.0
2007-01-29	2.0

至

date	pos
2007-01-01	NA
2007-01-02	NA
2007-01-03	NA
2007-01-10	11.0
2007-01-11	11.0
2007-01-20	11.0
2007-01-21	11.0
2007-01-22	13.0
2007-01-23	13.0
2007-01-27	13.0
2007-01-28	13.0
2007-01-29	13.0

我知道如何分别进行后滚和前滚。但是想不通怎么同时做。

Answer 1

您可以使用 np.where 将不符合您条件的值替换为 nan，然后使用 ffill 向前转换先前的值。

在这种情况下，我们对第一个条件使用扩展 window，最小周期为 4，对于第二个条件，我们反转数据并滚动 window 2。

import pandas as pd
import numpy as np

df = pd.DataFrame({'date': {0: '2007-01-01',
  1: '2007-01-02', 2: '2007-01-03', 3: '2007-01-10', 4: '2007-01-11',
  5: '2007-01-20', 6: '2007-01-21', 7: '2007-01-22', 8: '2007-01-23',
  9: '2007-01-27', 10: '2007-01-28', 11: '2007-01-29'},
 'pos': {0: 3.0, 1: 5.0, 2: 6.0, 3: 11.0, 4: 9.0, 5: 8.0,
  6: 10.0, 7: 13.0, 8: 4.0, 9: 2.0, 10: 1.0, 11: 2.0}})

df.pos = np.where((df.pos.ge(df.pos.rolling(len(df), min_periods=4).max())) & 
         (df.pos.ge(df.iloc[::-1].pos.rolling(2).max())),
        df.pos,np.nan)
df.ffill()

输出

    date    pos
0   2007-01-01  NaN
1   2007-01-02  NaN
2   2007-01-03  NaN
3   2007-01-10  11.0
4   2007-01-11  11.0
5   2007-01-20  11.0
6   2007-01-21  11.0
7   2007-01-22  13.0
8   2007-01-23  13.0
9   2007-01-27  13.0
10  2007-01-28  13.0
11  2007-01-29  13.0

如何在pandas中同时进行前后滚动？

how to do forward and backward rolling at the same time in pandas?

python

python-3.x

pandas

rolling-computation