使用 Pandas 数据帧逐行查找过去 6 个月内发生的事件

Question

假设我有一个这样的数据集：

  id_police id_sinistre    datesurv
0      p123        s120  01/01/2018
1      p123        s121  03/01/2018
2      p123        s122  05/05/2018
3      p222        s123  04/05/2018
4      p222        s124  02/12/2018
5      p433        s125  07/08/2018
6      p433        s126  08/09/2018
7      p433        s127  10/10/2018

我的目标是为每一行找到过去 6 个月内id_police的最后出现次数，如下所示：

  id_police id_sinistre    datesurv  occ
0      p123        s120  01/01/2018    0
1      p123        s121  03/01/2018    1
2      p123        s122  05/05/2018    2
3      p222        s123  04/05/2018    0
4      p222        s124  02/12/2018    0
5      p433        s125  07/08/2018    0
6      p433        s126  08/09/2018    1
7      p433        s127  10/10/2018    2

我想我需要.duplicated或.groupby但我不确定如何使用它们......提前感谢您的帮助！

Answer 1

如果 6 个月应该简化为 6 * 30 天，请使用自定义 lambda 函数和 diff ，按值和上次累积总和进行比较：

df['datesurv'] = pd.to_datetime(df['datesurv'], dayfirst=True)
df = df.sort_values(['id_police','datesurv'])
f = lambda x: (x.diff().dt.days < 30 * 6).cumsum()
df['occ'] = df.groupby('id_police')['datesurv'].apply(f)
print (df)
  id_police id_sinistre   datesurv  occ
0      p123        s120 2018-01-01    0
1      p123        s121 2018-01-03    1
2      p123        s122 2018-05-05    2
3      p222        s123 2018-05-04    0
4      p222        s124 2018-12-02    0
5      p433        s125 2018-08-07    0
6      p433        s126 2018-09-08    1
7      p433        s127 2018-10-10    2

Answer 2

另一种

选择是GroupBy datesurv，并使用pd.Grouper创建6个月的小组并采取cumcount：

df.datesurv = pd.to_datetime(df.datesurv, format='%d/%m/%Y')
g = pd.Grouper(key='datesurv', freq='6MS')
df.assign(occ=df.groupby(['id_police', g]).cumcount())
   id_police id_sinistre   datesurv  occ
0      p123        s120 2018-01-01    0
1      p123        s121 2018-01-03    1
2      p123        s122 2018-05-05    2
3      p222        s123 2018-05-04    0
4      p222        s124 2018-12-02    0
5      p433        s125 2018-08-07    0
6      p433        s126 2018-09-08    1
7      p433        s127 2018-10-10    2