过滤和扩展时间序列 pandas 数据框

Question

这个问题是这个问题的补充：filter multi-indexed grouped pandas dataframe

我想得到timestamp，其中value在date之后开始大于零，作为每个人id的新列new_date

示例输入数据：

id timestamp  date       value
1  2001-01-01 2001-05-01 1
1  2001-10-01 2001-05-01 0
1  2001-10-02 2001-05-01 1
1  2001-10-03 2001-05-01 0
1  2001-10-04 2001-05-01 1
2  2001-01-01 2001-05-01 1
2  2001-10-01 2001-05-01 0
2  2001-10-02 2001-05-01 0
2  2001-10-03 2001-05-01 0
2  2001-10-04 2001-05-01 1

想要的输出数据示例：

id timestamp  date       value new_date
1  2001-01-01 2001-05-01 1     2001-10-02
1  2001-10-01 2001-05-01 0     2001-10-02
1  2001-10-02 2001-05-01 1     2001-10-02
1  2001-10-03 2001-05-01 0     2001-10-02
1  2001-10-04 2001-05-01 1     2001-10-02
2  2001-01-01 2001-05-01 1     2001-10-04
2  2001-10-01 2001-05-01 0     2001-10-04
2  2001-10-02 2001-05-01 0     2001-10-04
2  2001-10-03 2001-05-01 0     2001-10-04
2  2001-10-04 2001-05-01 1     2001-10-04

Answer 1

如果某些组没有匹配项，也可以使用更简单的解决方案，首先过滤 DataFrame 链式掩码以获得更大的类似 Series.gt 的 date 和按位 AND 与 0 相同，然后通过 DataFrame.drop_duplicates 删除重复项，创建Series，最后使用Series.map：

df['timestamp'] = pd.to_datetime(df['timestamp'])
df['date'] = pd.to_datetime(df['date'])
df = df.sort_values(['id','timestamp'])

m = df['timestamp'].gt(df['date']) & df['value'].gt(0)

s = df[m].drop_duplicates('id').set_index('id')['timestamp']

df['new_date'] = df['id'].map(s)
print (df)
   id  timestamp       date  value   new_date
0   1 2001-01-01 2001-05-01      1 2001-10-02
1   1 2001-10-01 2001-05-01      0 2001-10-02
2   1 2001-10-02 2001-05-01      1 2001-10-02
3   1 2001-10-03 2001-05-01      0 2001-10-02
4   1 2001-10-04 2001-05-01      1 2001-10-02
5   2 2001-01-01 2001-05-01      1 2001-10-04
6   2 2001-10-01 2001-05-01      0 2001-10-04
7   2 2001-10-02 2001-05-01      0 2001-10-04
8   2 2001-10-03 2001-05-01      0 2001-10-04
9   2 2001-10-04 2001-05-01      1 2001-10-04