Python 多列go 除连续重复项

我有一些包含多列的时间序列数据:

         date hardware location  group   rtype  value
0 2021-03-01     desk       NY  opera  type-s      0
1 2021-03-01     desk       NJ  opera  type-s    200
2 2021-03-01     desk       IL  opera  type-s    100
3 2022-08-01     desk       NY  opera  type-s    275
4 2021-08-25     desk       IL  opera  type-s    100
5 2022-09-16     desk       IL  opera  type-s     30
6 2022-09-16     desk       NY  opera  type-s      0
7 2022-11-01     desk       NJ  opera  type-s      0
8 2022-11-01     desk       IL  opera  type-s      0

我想删除忽略日期的连续重复项，例如

...
2 2021-03-01     desk       IL  opera  type-s    100
3 2022-08-01     desk       NY  opera  type-s    275
4 2021-08-25     desk       IL  opera  type-s    100
...

索引为2、4的行是重复的，我只想保留第一行.

我试了两种方法，

1. 使用也丢弃非连续副本的drop_duplicates， 和
2. 使用shift() != ，我找到了in this answer个，似乎也在做同样的事情.

以下是我的测试代码:

import pandas as pd
from io import StringIO

data = """
date,hardware,location,group,rtype,value
2021-03-01,desk,NY,opera,type-s,0
2021-03-01,desk,NJ,opera,type-s,200
2021-03-01,desk,IL,opera,type-s,100
2022-08-01,desk,NY,opera,type-s,275
2021-08-25,desk,IL,opera,type-s,100
2022-09-16,desk,IL,opera,type-s,30
2022-09-16,desk,NY,opera,type-s,0
2022-11-01,desk,NJ,opera,type-s,0
2022-11-01,desk,IL,opera,type-s,0
"""

df = pd.read_csv(StringIO(data), parse_dates=['date'])

print('\nOriginal \n', df.to_string())

columns = ['hardware', 'location', 'group', 'rtype']


dedup_cols = columns + ['value']

df = df.drop_duplicates(
    subset=dedup_cols, keep='first').reset_index(drop=True)

de_dup = df.loc[(df[dedup_cols].shift() != df[dedup_cols]).any(axis=1)]

print('\nDrop_duplicates: \n', df.to_string())
print('\nShift_filter: \n', de_dup.to_string())

expected = """
date,hardware,location,group,rtype,value
2021-03-01,desk,NY,opera,type-s,0
2021-03-01,desk,NJ,opera,type-s,200
2021-03-01,desk,IL,opera,type-s,100
2022-08-01,desk,NY,opera,type-s,275
2022-09-16,desk,IL,opera,type-s,30
2022-09-16,desk,NY,opera,type-s,0
2022-11-01,desk,NJ,opera,type-s,0
2022-11-01,desk,IL,opera,type-s,0
"""

expected_df = pd.read_csv(StringIO(expected), parse_dates=['date'])

print('Equal drop_duplicates ? ', expected_df.equals(df))
print('Equal shift_filter ? ', expected_df.equals(de_dup))
print('Same result from the two techniques? ', de_dup.equals(df))
print('\nExpected:\n', expected_df.to_string())

-你在做什么？

Original 
         date hardware location  group   rtype  value
0 2021-03-01     desk       NY  opera  type-s      0
1 2021-03-01     desk       NJ  opera  type-s    200
2 2021-03-01     desk       IL  opera  type-s    100
3 2022-08-01     desk       NY  opera  type-s    275
4 2021-08-25     desk       IL  opera  type-s    100
5 2022-09-16     desk       IL  opera  type-s     30
6 2022-09-16     desk       NY  opera  type-s      0
7 2022-11-01     desk       NJ  opera  type-s      0
8 2022-11-01     desk       IL  opera  type-s      0

Drop_duplicates: 
         date hardware location  group   rtype  value
0 2021-03-01     desk       NY  opera  type-s      0
1 2021-03-01     desk       NJ  opera  type-s    200
2 2021-03-01     desk       IL  opera  type-s    100
3 2022-08-01     desk       NY  opera  type-s    275
4 2022-09-16     desk       IL  opera  type-s     30
5 2022-11-01     desk       NJ  opera  type-s      0
6 2022-11-01     desk       IL  opera  type-s      0

Shift_filter: 
         date hardware location  group   rtype  value
0 2021-03-01     desk       NY  opera  type-s      0
1 2021-03-01     desk       NJ  opera  type-s    200
2 2021-03-01     desk       IL  opera  type-s    100
3 2022-08-01     desk       NY  opera  type-s    275
4 2022-09-16     desk       IL  opera  type-s     30
5 2022-11-01     desk       NJ  opera  type-s      0
6 2022-11-01     desk       IL  opera  type-s      0
Equal drop_duplicates ?  False
Equal shift_filter ?  False
Same result from the two techniques?  True

Expected:
         date hardware location  group   rtype  value
0 2021-03-01     desk       NY  opera  type-s      0
1 2021-03-01     desk       NJ  opera  type-s    200
2 2021-03-01     desk       IL  opera  type-s    100
3 2022-08-01     desk       NY  opera  type-s    275
4 2022-09-16     desk       IL  opera  type-s     30
5 2022-09-16     desk       NY  opera  type-s      0
6 2022-11-01     desk       NJ  opera  type-s      0
7 2022-11-01     desk       IL  opera  type-s      0

我认为drop_duplicates的方法是不合适的.

我希望第二种方法能行得通，但我不完全明白为什么行不通.

我在想一些group_by+shift的方法，然后取一个连续的差值，然后在差值为0的地方下降.有什么主意吗？