years-only in date_range uses current date

[rosnfeld]

I ran this on master current as of 2014-04-20 or so and was surprised as to the result:

In [17]: pd.date_range('2014', '2015', freq='M')
Out[17]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2014-04-30, ..., 2015-03-31]
Length: 12, Freq: M, Timezone: None

I would have expected that to give me the month-ends in 2014, not a year's worth of month-ends starting from today.

Seems to be related to #3944?

[jreback]

I think this is simply an ambiguity, you can (and should specify the start or end date more fully)

e.g. 20140101

maybe could somehow provide a better inference on the default date though if its a 'partial' date.
feel free to look into it (but I think its the same issue as #3944), so going to close this and copy their

[jreback]

sorry..realized the other issue is closed (sort of punted on it). as I said, feel free to try to address.

[sinhrks]

Going to be covered in #7599.

Result after #7599

The problem is caused by a path which uses today's date as default. Being fixed to use first day of the year (and specifying freq outputs expected result).

pd.date_range('2014', '2015', freq='M')
# <class 'pandas.tseries.index.DatetimeIndex'>
# [2014-01-31, ..., 2014-12-31]
# Length: 12, Freq: M, Timezone: None

pd.date_range('2014', '2015', freq='MS')
# <class 'pandas.tseries.index.DatetimeIndex'>
# [2014-01-01, ..., 2015-01-01]
# Length: 13, Freq: MS, Timezone: None

Other datetime parse function results (for ref)

Single year is parsed as the first day of the year in most function already (not changed by #7599).

pd.tseries.tools.parse_time_string('2014')[0]
# 2014-01-01 00:00:00

pd.tseries.tools.to_datetime('2014')
# 2014-01-01 00:00:00

[rosnfeld]

Great! I did some ad-hoc testing on your PR and confirm it works as expected. Thanks!

[jreback]

@sinhrks this also seems to be fixed in current master ? (irrespective of #7599)

[sinhrks]

Ah correct. Maybe caused by #7907 replaces dateutil.parse by _string_to_dts.

Adding an explicit test in #7599 though...

[jreback]

ok, so in #7599 just fix up the release notes if you would

[rosnfeld]

As this seems to be fixed in 0.15.0, should we just close this issue?

[jreback]

really? can u a) see if u can find a refence to an issue which closed this ?

if it works can u put a test that validates?

[rosnfeld]

Yes, you were actually the first person to figure this out, in the comments above. I'm guessing it's #7907 that fixed it, as @sinhrks said. I can add a test though.

[jreback]

ahh ok
I think that fixed s big that had it go to dateutil which does the wrong thing (and thus it work)
can u add s test release note and well call it a day
thxs