如何通过比较具有 NaN 值的两列来获得最大的日期

Question

1.我有一个 DF，其中日期应始终大于 cop_date 和 fat_date。

2.如果fact是nan，它仍然应该获取大于cop_date的日期

3.只考虑最近的较小日期。

实际DF：

id  ins_id  date    cop cop_date    fat fat_date
1234    abc 17/04/2023  1   18/05/2023  nan nan
1234    abc 16/04/2023  1   12/03/2023  nan nan
1234    abc 18/04/2023  1   23/03/2023  nan nan
1234    ghi 22/06/2023  1   27/08/2023  2   15/09/2023
1234    ghi 23/06/2023  1   22/05/2023  2   20/10/2023
1234    ghi 26/06/2023  1   19/04/2023  2   30/04/2023
1234    jkl 22/08/2023  1   26/08/2023  nan nan
1234    jkl 17/08/0223  1   13/08/2023  nan nan
1234    mno 06/05/2023  1   09/05/2023  2   10/05/2023
1234    mno 04/05/2023  1   01/05/2023  2   01/04/2023

预期DF：

id  ins_id  date    cop cop_date    fat fat_date
1234    abc 17/04/2023  1   12/03/2023  nan nan
1234    abc 16/04/2023  1   12/03/2023  nan nan
1234    abc 18/04/2023  1   12/03/2023  nan nan
1234    ghi 22/06/2023  1   19/04/2023  2   30/04/2023
1234    ghi 23/06/2023  1   19/04/2023  2   30/04/2023
1234    ghi 26/06/2023  1   19/04/2023  2   30/04/2023
1234    jkl 22/08/2023  1   13/08/2023  nan nan
1234    jkl 17/08/0223  1   13/08/2023  nan nan
1234    mno 06/05/2023  1   01/05/2023  2   01/04/2023
1234    mno 04/05/2023  1   01/05/2023  2   01/04/2023

我尝试过的查询：

Data =DF[(DF[`date`] > DF[`cop_date`]) & (DF[`date`] > DF[`fat_date`])]

Data= data.drop_duplicates(subset=['id','ins_id','date'],keep='first')

上面的查询删除了具有 nan 值的行。

Answer 1

用途：

cols = ['id','ins_id']

df['cop_date'] = df.assign(d = df['cop_date'].where(df['cop_date'].lt(df['date']))).groupby(cols)['d'].transform('min')
df['fat_date'] = df.assign(d = df['fat_date'].where(df['fat_date'].lt(df['date']))).groupby(cols)['d'].transform('min')

print (df)
     id ins_id       date  cop   cop_date  fat   fat_date
0  1234    abc 2023-04-17    1 2023-03-12  NaN        NaT
1  1234    abc 2023-04-16    1 2023-03-12  NaN        NaT
2  1234    abc 2023-04-18    1 2023-03-12  NaN        NaT
3  1234    ghi 2023-06-22    1 2023-04-19  2.0 2023-04-30
4  1234    ghi 2023-06-23    1 2023-04-19  2.0 2023-04-30
5  1234    ghi 2023-06-26    1 2023-04-19  2.0 2023-04-30
6  1234    jkl 2023-08-22    1 2023-08-13  NaN        NaT
7  1234    jkl 2023-08-17    1 2023-08-13  NaN        NaT
8  1234    mno 2023-05-06    1 2023-05-01  2.0 2023-04-01
9  1234    mno 2023-05-04    1 2023-05-01  2.0 2023-04-01