API: DatetimeIndex / TimedeltaIndex 2d slicing should result in Index

[sinhrks]

Following slices basically result in the same Index.

import pandas as pd
pd.Index([1, 2, 3])[:, None]
# Int64Index([[1], [2], [3]], dtype='int64')
pd.MultiIndex.from_tuples([('a', 1), ('b', 2)])[:, None]
# MultiIndex(levels=[[u'a', u'b'], [1, 2]],
#            labels=[[[0], [1]], [[0], [1]]])

pd.period_range('2011-01-01', freq='M', periods=3)[:, None]
# PeriodIndex(['2011-01', '2011-02', '2011-03'], dtype='int64', freq='M')

But DatetimeIndex and TimedeltaIndex doesn't because of following line.

• https://github.com/pydata/pandas/blob/master/pandas/tseries/base.py#L119

pd.date_range('2011-01-01', freq='M', periods=3)[:, None]
# array([['2011-01-31T09:00:00.000000000+0900'],
#        ['2011-02-28T09:00:00.000000000+0900'],
#        ['2011-03-31T09:00:00.000000000+0900']], dtype='datetime64[ns]')
pd.timedelta_range('1 days', freq='D', periods=3)[:, None]
# array([[ 86400000000000],
#        [172800000000000],
#        [259200000000000]], dtype='timedelta64[ns]')

I understand these must be Index also, or any reason for this?

[jreback]

prob an oversite

but not sure we should allow 2-d results at all
it's very odd (and I don't any utility)

[sinhrks]

Yep. Raise IndexError as the same as explicit 2 indices?

pd.Index([1, 2, 3])[:, 0]
# IndexError: too many indices for array

[jreback]

sounds right

[shoyer]

I noticed this the other day. I agree with @jreback that we should raise an IndexError if the result has two or more dimensions.

[ShaharNaveh]

@jreback Since #31150 this is raising a Deprecation warning

pandas/pandas/core/indexers.py

Lines 263 to 270 in b1214af

	if np.ndim(result) > 1:
	warnings.warn(
	"Support for multi-dimensional indexing (e.g. `index[:, None]`) "
	"on an Index is deprecated and will be removed in a future "
	"version. Convert to a numpy array before indexing instead.",
	DeprecationWarning,
	stacklevel=3,
	)

---

Should it raise an IndexError instead of a deprecation warning? (as was agreed above)