如何用Pandas获取DataFrame中连续NaN区间的high/low极值？

我正在处理如下结构的DataFrame：

datetime     high      low
1   2022-08-26 19:00:00+00:00      NaN  0.99564
6   2022-08-26 14:00:00+00:00  1.00902      NaN
9   2022-08-26 11:00:00+00:00      NaN  0.99860
10  2022-08-26 10:00:00+00:00  1.00238      NaN
14  2022-08-26 06:00:00+00:00      NaN  0.99466
17  2022-08-26 03:00:00+00:00  0.99748      NaN
22  2022-08-25 22:00:00+00:00  0.99772      NaN
25  2022-08-25 19:00:00+00:00  0.99790      NaN
27  2022-08-25 17:00:00+00:00  0.99752      NaN
28  2022-08-25 16:00:00+00:00      NaN  0.99492
30  2022-08-25 14:00:00+00:00  1.00006      NaN
31  2022-08-25 13:00:00+00:00      NaN  0.99555
38  2022-08-25 06:00:00+00:00  1.00336      NaN
40  2022-08-25 04:00:00+00:00  1.00088      NaN
43  2022-08-25 01:00:00+00:00  0.99929      NaN
44  2022-08-25 00:00:00+00:00      NaN  0.99632
47  2022-08-24 21:00:00+00:00      NaN  0.99636

期望得到如下输出：

datetime     high      low
1   2022-08-26 19:00:00+00:00      NaN  0.99564
6   2022-08-26 14:00:00+00:00  1.00902      NaN
9   2022-08-26 11:00:00+00:00      NaN  0.99860
10  2022-08-26 10:00:00+00:00  1.00238      NaN
14  2022-08-26 06:00:00+00:00      NaN  0.99466
25  2022-08-25 19:00:00+00:00  0.99790      NaN
28  2022-08-25 16:00:00+00:00      NaN  0.99492
30  2022-08-25 14:00:00+00:00  1.00006      NaN
31  2022-08-25 13:00:00+00:00      NaN  0.99555
38  2022-08-25 06:00:00+00:00  1.00336      NaN
44  2022-08-25 00:00:00+00:00      NaN  0.99632

具体需求：针对每一段连续的high为NaN（对应取low的最小值）或low为NaN（对应取high的最大值）的区间，获取该区间内的极值行，而非全局极值。想找一种简洁的Pandas实现方式，替代分块处理的方法。

可以通过Pandas的分组功能，结合连续区间标记实现，步骤如下：

1. 标记连续区间分组键
   先区分每行的类型（high非空/low非空），再通过类型变化生成连续区间的分组ID：

   # 标记类型：1代表high非空（low为空），0代表low非空（high为空）
   df['group_type'] = df['high'].notna().astype(int)
   # 生成连续区间的分组ID：类型变化时ID递增
   df['group_id'] = (df['group_type'] != df['group_type'].shift()).cumsum()
   

2. 分组提取极值行
   针对每个分组，根据类型提取对应列的极值所在行：

   def extract_extreme(group):
       if group['group_type'].iloc[0] == 1:
           # high非空组，取high最大值对应的行
           return group[group['high'] == group['high'].max()]
       else:
           # low非空组，取low最小值对应的行
           return group[group['low'] == group['low'].min()]
   
   # 应用函数并合并结果，移除辅助列
   result = df.groupby('group_id').apply(extract_extreme).reset_index(drop=True)
   result = result.drop(['group_type', 'group_id'], axis=1)
   

运行上述代码后，result即为期望的输出结果。

内容的提问来源于stack exchange，提问作者som3k