PeriodIndex with float input inconsistency

[sinhrks]

Code Sample, a copy-pastable example if possible

pd.PeriodIndex([1.1], freq='M')
# ValueError: Given date string not likely a datetime.

pd.PeriodIndex(np.array([1.1]), freq='M')
# PeriodIndex(['1970-02'], dtype='int64', freq='M'

Expected Output

Consistently raise or coerce to Period

output of pd.show_versions()

0.18.1

CC: @MaximilianR

[max-sixty]

ref #13079, which should fix

[max-sixty]

@sinhrks @jreback closed by #13079

[jreback]

@MaximilianR if you think we need to add a note (and/or include these issues) in the whatsnew, pls do a PR. I don't care about the issues per-se being referenced by GH does that. But if you are a user and want to know (and the existing notes don't cover), then pls add.

[max-sixty]

@jreback this seems like pretty deep in the internals - shall I still add one?

[jreback]

nah I looked again, its fine. thxs!

Periods really coming along!

[sinhrks]

@MaximilianR #13079 only checks _shallow_copy with normal float input right?

on current master (d2b5819):

pd.PeriodIndex(np.array([1.1, np.nan, np.inf]), freq='M')
# PeriodIndex(['1970-02', 'NaT', 'NaT'], dtype='int64', freq='M')

pd.PeriodIndex([1.1, np.nan, np.inf], freq='M')
# ValueError: Value must be Period, string, integer, or datetime

reopens.

[jreback]

In [2]: pd.PeriodIndex(np.array([1.1]), freq='M')
Out[2]: PeriodIndex(['1970-02'], dtype='int64', freq='M')

actually the original [2] should be an error as well (and ValueError is correct here)

[max-sixty]

Apologies. Will PR

[max-sixty]

Why a ValueError rather than TypeError?