我重新创建了你的dataFrame并尝试获取你的输出。我认为你可能在根据条件进行过滤之前错过了按标签对数据集进行排序。条件看起来很简单，只是不要忘记最后的按位非操作。你得到的输出只是相反的。

import pandas as pd

data = {
    'first_date': ['2023-09-11', '2023-09-11', '2023-09-20', '2023-09-17', '2023-09-30', '2023-10-03', '2023-10-05', '2023-10-20', '2023-10-10'],
    'label': ['A', 'B', 'A', 'B', 'A', 'A', 'B', 'A', 'B'],
    'last_date': ['2023-09-17', '2023-09-15', '2023-09-28', '2023-09-30', '2023-10-05', '2023-10-07', '2023-10-15', '2023-10-19', '2023-10-15']
}

df = pd.DataFrame(data)
df['first_date'] = pd.to_datetime(df['first_date'])
df['last_date'] = pd.to_datetime(df['last_date'])
print("Original DataFrame")
print(df)

# I guess you'were missing this part
df = df.sort_values(by=['label'])

# put your condition to check if first_date in between 
# first_date and last_date of any row above row z of the same label 

mask = []
for label, group in df.groupby('label'):
    date_filter = ~((group['first_date'] > group['first_date'].shift().fillna(pd.to_datetime('1900-01-01'))) & 
                  (group['first_date'] < group['last_date'].shift().fillna(pd.to_datetime('1900-01-01'))))
    mask.extend(date_filter.tolist())

filtered_df = df[mask]
print("\nFiltered DataFrame: ")
print(filtered_df)

字符串
输出量：

Original DataFrame
  first_date label  last_date
0 2023-09-11     A 2023-09-17
1 2023-09-11     B 2023-09-15
2 2023-09-20     A 2023-09-28
3 2023-09-17     B 2023-09-30
4 2023-09-30     A 2023-10-05
5 2023-10-03     A 2023-10-07
6 2023-10-05     B 2023-10-15
7 2023-10-20     A 2023-10-19
8 2023-10-10     B 2023-10-15

Filtered DataFrame: 
  first_date label  last_date
0 2023-09-11     A 2023-09-17
2 2023-09-20     A 2023-09-28
4 2023-09-30     A 2023-10-05
7 2023-10-20     A 2023-10-19
1 2023-09-11     B 2023-09-15
3 2023-09-17     B 2023-09-30
6 2023-10-05     B 2023-10-15

型

不幸的是，没有有效的方法在纯Pandas中做到这一点。
但是，您可以使用janitor的高效conditional_join执行左合并并识别匹配行，然后使用右索引删除它们：

# pip install pyjanitor
import janitor

df[['first_date', 'last_date']] = df[['first_date', 'last_date']].apply(pd.to_datetime)

drop = (df.reset_index()
          .conditional_join(df.reset_index(), 
                            ('first_date', 'first_date', '>='),
                            ('first_date', 'last_date', '<='),
                            ('label', 'label', '=='),
                            ('index', 'index', '>'),
                            how='left')
         .dropna(subset=[('right', 'index')])['left']['index'].unique()
       )

out = df.drop(drop)

字符串

• 注意，假设这里有一个范围索引，如果没有，就给列index分配一个范围，而不是使用reset_index。

输出量：

first_date label  last_date
0 2023-09-11     A 2023-09-17
1 2023-09-11     B 2023-09-15
2 2023-09-20     A 2023-09-28
3 2023-09-17     B 2023-09-30
4 2023-09-30     A 2023-10-05
5 2023-10-03     A 2023-10-07
6 2023-10-05     B 2023-10-15
7 2023-10-19     A 2023-10-20
8 2023-10-10     B 2023-10-15

型
中间体drop：array([5, 8])
合并中间体：

left                             right                            
  index first_date label  last_date index first_date label  last_date
0     0 2023-09-11     A 2023-09-17   NaN        NaT   NaN        NaT
1     1 2023-09-11     B 2023-09-15   NaN        NaT   NaN        NaT
2     2 2023-09-20     A 2023-09-28   NaN        NaT   NaN        NaT
3     3 2023-09-17     B 2023-09-30   NaN        NaT   NaN        NaT
4     4 2023-09-30     A 2023-10-05   NaN        NaT   NaN        NaT
5     5 2023-10-03     A 2023-10-07   4.0 2023-09-30     A 2023-10-05
6     6 2023-10-05     B 2023-10-15   NaN        NaT   NaN        NaT
7     7 2023-10-19     A 2023-10-20   NaN        NaT   NaN        NaT
8     8 2023-10-10     B 2023-10-15   6.0 2023-10-05     B 2023-10-15

型

我会对这个框架进行排序，然后使用Apply和一个记住最后一行处理的函数来覆盖它：

• https://trinket.io/python3/43f1a6d2f4

import pandas as pd
import numpy as np

data = [
  ['2023-09-11', 'A', '2023-09-17'],
  ['2023-09-11', 'B', '2023-09-15'],
  ['2023-09-20', 'A', '2023-09-28'],
  ['2023-09-17', 'B', '2023-09-30'],
  ['2023-09-30', 'A', '2023-10-05'],
  ['2023-10-03', 'A', '2023-10-07'],
  ['2023-10-05', 'B', '2023-10-15'],
  ['2023-10-19', 'A', '2023-10-20'],
  ['2023-10-10', 'B', '2023-10-15'],
  ['2023-10-12', 'B', '2023-10-18'], # Added by me
  ['2023-10-16', 'B', '2023-10-20'], # Added by me
]

df = pd.DataFrame(
  data,
  columns = ['first_date', 'label', 'last_date'],
)

df['first_date'] = pd.to_datetime(df['first_date'])
df['last_date' ] = pd.to_datetime(df['last_date' ])

print(df.dtypes)
print()

df.sort_values(['label', 'first_date', 'last_date'], inplace = True)
print(df)
print()

class Filter:
  def __init__(self):
    self._label = None
    self._first = None
    self._last  = None
  
  def _persist(self, row):
    self._label = row['label']
    self._first = row['first_date']
    self._last  = row['last_date']
  
  def validate(self, row):
    if self._label != row['label']:
      self._persist(row)
      return True
    
    if self._first <= row['first_date'] <= self._last:
      return False
    
    self._persist(row)
    return True
    

filter = Filter()

df['flag'] = df.apply(filter.validate, axis=1)

print(df)
print()

first_date label  last_date   flag
0  2023-09-11     A 2023-09-17   True
2  2023-09-20     A 2023-09-28   True
4  2023-09-30     A 2023-10-05   True
5  2023-10-03     A 2023-10-07  False
7  2023-10-19     A 2023-10-20   True
1  2023-09-11     B 2023-09-15   True
3  2023-09-17     B 2023-09-30   True
6  2023-10-05     B 2023-10-15   True
8  2023-10-10     B 2023-10-15  False
9  2023-10-12     B 2023-10-18  False
10 2023-10-16     B 2023-10-20   True

import numpy as np
# merge on self (highly used sql trick)
new_df = df.merge(df, on="label", how="left",  suffixes=("","_new"))
# filter similar dates (not a necessary step)
new_df = new_df[(new_df["first_date"] != new_df["first_date_new"]) ]
# flag where your conditions is true
new_df["flag"] = np.where((new_df["first_date_new"] < new_df["first_date"]) & (new_df["first_date"] < new_df["last_date_new"]), True,False )

to_drop = new_df[new_df["flag"]].copy()

to_drop["shift_first_date_new"] = to_drop["first_date"].shift(+1)
to_drop["flag"] = to_drop["flag"] & (to_drop["first_date_new"] != to_drop["shift_first_date_new"])

to_drop = to_drop[to_drop["flag"]][["first_date","label","last_date"]]
pd.concat([df,to_drop]).drop_duplicates(keep=False).sort_values("label")
#
  first_date label  last_date
0 2023-09-11     A 2023-09-17
1 2023-09-11     B 2023-09-15
2 2023-09-20     A 2023-09-28
3 2023-09-17     B 2023-09-30
4 2023-09-30     A 2023-10-05
6 2023-10-05     B 2023-10-15
7 2023-10-19     A 2023-10-20

    first_date label   last_date
0   2023-09-11     A  2023-09-17
2   2023-09-20     A  2023-09-28
4   2023-09-30     A  2023-10-05
7   2023-10-19     A  2023-10-20
1   2023-09-11     B  2023-09-15
3   2023-09-17     B  2023-09-30
6   2023-10-05     B  2023-10-15
10  2023-10-16     B  2023-10-20

字符串