一个选项是交换 * 前缀/后缀 *，以便为wide_to_long整形创建 stubnames：

out = (df.set_index(["id", "university"])
         .pipe(lambda x: x.set_axis([f"{c2}_{c1}"
               for _,c1,c2 in x.columns.str.split(r"\s|_")], axis=1)).reset_index()
         .pipe(pd.wide_to_long, stubnames=["Holiday", "Hours"],
                  i=["id", "university"], j="Holiday_State", sep="_", suffix=r"\w+")
         .reset_index()
      )

输出：

print(out)
           
   id university Holiday_State  Holiday  Hours
0   1        abc            CA     1.00   5.00
1   1        abc            WA      NaN   3.00
2   2        def            CA     5.00    NaN
3   2        def            WA     3.00   7.00
4   3        ijk            CA     1.00   5.00
5   3        ijk            WA      NaN   3.00

一个选项是使用pivot_longer：

# pip install pyjanitor
import pandas as pd
import janitor

df.pivot_longer(index = ['id', 'university'], 
                names_to = ['Holiday_State', '.value'], 
                names_pattern = r".\s([a-zA-Z]+).(.+)",
                sort_by_appearance = True)
   id university Holiday_State  Holiday  Hours
0   1        abc            CA      1.0    5.0
1   1        abc            WA      NaN    3.0
2   2        def            CA      5.0    NaN
3   2        def            WA      3.0    7.0
4   3        ijk            CA      1.0    5.0
5   3        ijk            WA      NaN    3.0