[Proposal] Allow open boundaries when some of them are undefined for DatetimeIndex

[Adirio]

Problem description

When creating a DatetimeIndex partially-bounded, i.e., only specifying either start or end the use of a closed parameter different from None was intended to be disallowed. This behaviour is bugged as shown in #23198. #23199 solves this bug.

Before #23199, the DatetimeIndex was successfully created. The result contained one less than the requested periods in some cases as shown in #23176.

After #23199, a ValueError will be raised as it seemed to be intended.

The proposal is to allow closed parameters different to None for partially-bounded DatetimeIndexes. The behaviour should be as defined in the following table

Periods	Ommited parameter	Closed	Current number of items	After #23199	Proposed number of items
N	end	None	N	N	N
N	end	"left"	N-1	ValueError	N
N	end	"right"	N	ValueError	N
N	start	None	N	N	N
N	start	"left"	N	ValueError	N
N	start	"right"	N-1	ValueError	N

[jbrockmendel]

Can you give copy/pasteable examples of what behavior you want to change and what you want it changed to?

[Adirio]

Note: this issue refers to the arguments start, periods, end and closed which were deprecated in version 0.24.0 and are not present in the latest pandas version (1.1.2). Is there still any version receiving bug fixes that still implements these arguments?

The rest of this comment is using pandas version 0.23.4

I expected the following commands to give 10 results but some don't:

from datetime import datetime, timedelta
import pandas as pd

n = 10
start = datetime.now()
end = start + timedelta(seconds=n-1)
pd.DatetimeIndex(freq='s', periods=n, start=start, closed=None)     # 1
pd.DatetimeIndex(freq='s', periods=n, start=start, closed='left')   # 2
pd.DatetimeIndex(freq='s', periods=n, start=start, closed='right')  # 3
pd.DatetimeIndex(freq='s', periods=n, end=end, closed=None)         # 4
pd.DatetimeIndex(freq='s', periods=n, end=end, closed='left')       # 5
pd.DatetimeIndex(freq='s', periods=n, end=end, closed='right')      # 6

Numbers 1, 2, 4 and 6 return:

DatetimeIndex(['2020-10-02 16:54:35.329660', '2020-10-02 16:54:36.329660',
               '2020-10-02 16:54:37.329660', '2020-10-02 16:54:38.329660',
               '2020-10-02 16:54:39.329660', '2020-10-02 16:54:40.329660',
               '2020-10-02 16:54:41.329660', '2020-10-02 16:54:42.329660',
               '2020-10-02 16:54:43.329660', '2020-10-02 16:54:44.329660'],
              dtype='datetime64[ns]', freq='S')

Number 3 returns:

DatetimeIndex(['2020-10-02 16:54:36.329660', '2020-10-02 16:54:37.329660',
               '2020-10-02 16:54:38.329660', '2020-10-02 16:54:39.329660',
               '2020-10-02 16:54:40.329660', '2020-10-02 16:54:41.329660',
               '2020-10-02 16:54:42.329660', '2020-10-02 16:54:43.329660',
               '2020-10-02 16:54:44.329660'],
              dtype='datetime64[ns]', freq='S')

Number 5 returns:

DatetimeIndex(['2020-10-02 16:54:35.329660', '2020-10-02 16:54:36.329660',
               '2020-10-02 16:54:37.329660', '2020-10-02 16:54:38.329660',
               '2020-10-02 16:54:39.329660', '2020-10-02 16:54:40.329660',
               '2020-10-02 16:54:41.329660', '2020-10-02 16:54:42.329660',
               '2020-10-02 16:54:43.329660'],
              dtype='datetime64[ns]', freq='S')

[jbrockmendel]

Is there still any version receiving bug fixes that still implements these arguments?

Use pd.date_range

[Adirio]

In version 1.1.2, pd.date_range has the exact same behavior as pd.DatetimeIndex in 0.23.4, so I guess this still aplies.

from datetime import datetime, timedelta
import pandas as pd

n = 10
start = datetime.now()
end = start + timedelta(seconds=n-1)
pd.date_range(freq='s', periods=n, start=start, closed=None)     # 1
pd.date_range(freq='s', periods=n, start=start, closed='left')   # 2
pd.date_range(freq='s', periods=n, start=start, closed='right')  # 3
pd.date_range(freq='s', periods=n, end=end, closed=None)         # 4
pd.date_range(freq='s', periods=n, end=end, closed='left')       # 5
pd.date_range(freq='s', periods=n, end=end, closed='right')      # 6

Numbers 1, 2, 4 and 6 return:

DatetimeIndex(['2020-10-02 16:54:35.329660', '2020-10-02 16:54:36.329660',
               '2020-10-02 16:54:37.329660', '2020-10-02 16:54:38.329660',
               '2020-10-02 16:54:39.329660', '2020-10-02 16:54:40.329660',
               '2020-10-02 16:54:41.329660', '2020-10-02 16:54:42.329660',
               '2020-10-02 16:54:43.329660', '2020-10-02 16:54:44.329660'],
              dtype='datetime64[ns]', freq='S')

Number 3 returns:

DatetimeIndex(['2020-10-02 16:54:36.329660', '2020-10-02 16:54:37.329660',
               '2020-10-02 16:54:38.329660', '2020-10-02 16:54:39.329660',
               '2020-10-02 16:54:40.329660', '2020-10-02 16:54:41.329660',
               '2020-10-02 16:54:42.329660', '2020-10-02 16:54:43.329660',
               '2020-10-02 16:54:44.329660'],
              dtype='datetime64[ns]', freq='S')

Number 5 returns:

DatetimeIndex(['2020-10-02 16:54:35.329660', '2020-10-02 16:54:36.329660',
               '2020-10-02 16:54:37.329660', '2020-10-02 16:54:38.329660',
               '2020-10-02 16:54:39.329660', '2020-10-02 16:54:40.329660',
               '2020-10-02 16:54:41.329660', '2020-10-02 16:54:42.329660',
               '2020-10-02 16:54:43.329660'],
              dtype='datetime64[ns]', freq='S')

[mroeschke]

Thanks for the request, but it appears this feature request hasn't gain much traction in years so closing