Python Pandas 数据框自定义排序功能

您可以使用口罩和numpy.lexsort:

import numpy as np

m = df['trigger'].isin(df.index)
key = df['trigger'].where(m, df.index)

out = df.iloc[np.lexsort([m, key])]

或者是纯Pandas (不那么优雅的IMO，但如果你需要定制订单，则更灵活；见下文):

out = (df
   .assign(m=df['trigger'].isin(df.index),
           key=lambda d: d['trigger'].where(m, d.index))
   .sort_values(by=['key', 'm'])
   .drop(columns=['m', 'key'])
)

输出:

     trigger   command
key1   cron1  command1
key2    key1  command2
key3   cron3  command3
key4    key3  command4
key5   cron5  command5

中间体:

# before sorting
     trigger   command      m   key
key4    key3  command4   True  key3
key2    key1  command2   True  key1
key3   cron3  command3  False  key3
key5   cron5  command5  False  key5
key1   cron1  command1  False  key1

# after sorting
     trigger   command      m   key
key1   cron1  command1  False  key1
key2    key1  command2   True  key1
key3   cron3  command3  False  key3
key4    key3  command4   True  key3
key5   cron5  command5  False  key5

如果要保持组的原始顺序(key3 -> key5 -> key1):

out = (df
   .assign(m=df['trigger'].isin(df.index),
           key=lambda d: pd.Categorical(d['trigger'].where(m, d.index),
                                       categories=d.index[~d['m']].unique(),
                                        ordered=True)
          )
   .sort_values(by=['key', 'm'])
   .drop(columns=['m', 'key'])
)

变体:

m = df['trigger'].isin(df.index)
key = df['trigger'].where(m, df.index)
order = df.index[~m].unique()

tmp = df.assign(key=pd.Categorical(key, categories=order, ordered=True))

out = (pd.concat([tmp[~m], tmp[m]]).sort_values(by='key', kind='stable')
         .drop(columns='key')
      )

输出:

     trigger   command
key3   cron3  command3
key4    key3  command4
key5   cron5  command5
key1   cron1  command1
key2    key1  command2