period_range creates wrong dates when freq has multiple offsets

[achabotl]

Code Sample, a copy-pastable example if possible

>>> import pandas as pd
>>> pd.__version__
u'0.18.0'
>>> pd.period_range('2016-07-20', periods=4, freq='2H')
PeriodIndex(['2016-07-20 00:00', '2016-07-20 02:00', '2016-07-20 04:00',
             '2016-07-20 06:00'],
            dtype='int64', freq='2H')
>>> pd.period_range('2016-07-20', periods=4, freq='2H30min')
PeriodIndex(['1970-10-11 08:48', '1970-10-11 08:50', '1970-10-11 08:52',
             '1970-10-11 08:54'],
            dtype='int64', freq='150T')
>>> pd.period_range('2016-07-20', periods=4, freq='2H30T')
PeriodIndex(['1970-10-11 08:48', '1970-10-11 08:50', '1970-10-11 08:52',
             '1970-10-11 08:54'],
            dtype='int64', freq='150T')
>>> pd.period_range('2016-07-20', periods=2, freq='1D1H')
PeriodIndex(['1971-12-10 10:00', '1971-12-10 11:00'], dtype='int64', freq='25H')

date_range works fine:

>>> pd.date_range('2016-07-20', periods=4, freq='2H30T')
DatetimeIndex(['2016-07-20 00:00:00', '2016-07-20 02:30:00',
               '2016-07-20 05:00:00', '2016-07-20 07:30:00'],
              dtype='datetime64[ns]', freq='150T')

Expected Output

The freq is right, but I'd expect the PeriodIndex to start at 2016-07-20, not in 1970, and to display 2h30 increments, something like:

PeriodIndex(['2016-07-20 00:00', '2016-07-20 02:30', '2016-07-20 05:00',
             '2016-07-20 07:30'],
            dtype='int64', freq='150T')

Or, if freq that combine multiple offsets are note supported, to at least raise and error.

output of pd.show_versions()

INSTALLED VERSIONS

---

commit: None
python: 2.7.11.final.0
python-bits: 64
OS: Darwin
OS-release: 15.5.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: en_US.UTF-8
LANG: en_US.UTF-8

pandas: 0.18.0
nose: 1.3.7
pip: 8.1.2
setuptools: 23.1.0
Cython: 0.24
numpy: 1.10.4
scipy: 0.17.1
statsmodels: 0.6.1
xarray: 0.7.2
IPython: 4.1.2
sphinx: 1.4.1
patsy: 0.4.1
dateutil: 2.5.2
pytz: 2016.3
blosc: 1.2.8
bottleneck: 1.1.0
tables: 3.2.2
numexpr: 2.6.0
matplotlib: 1.5.1
openpyxl: 2.3.4
xlrd: 0.9.4
xlwt: 1.0.0
xlsxwriter: 0.9.2
lxml: 3.6.0
bs4: 4.4.1
html5lib: 0.999
httplib2: 0.9.2
apiclient: None
sqlalchemy: 1.0.12
pymysql: 0.6.7.None
psycopg2: None
jinja2: 2.8
boto: 2.40.0

[jorisvandenbossche]

Confirmed on master, thanks for reporting!

[jorisvandenbossche]

For now, of course, you can use pd.period_range('2016-07-20', periods=4, freq='150T').

Before 0.17 this raised as multiples of a freq were not supported before (#7832). When adding that feature probably not checked for combinations.

cc @sinhrks

[sinhrks]

Yeah, Period can't support combinasions as it is under current impl. Should raise or coerce to single freq.

[jorisvandenbossche]

@sinhrks In some way it already does coerce the combination correctly to a single freq, as in the output of the examples you see 150T

[sinhrks]

@jorisvandenbossche yes "150T" should work after #7832. What I tried to mean is coercing "2H30T" to "150T" internally.

[jorisvandenbossche]

Yep, I understood that :-) What I meant is that somewhere this coercing already happens (only not in the right place for letting this work), since:

In [44]:  pd.period_range('2016-07-20', periods=4, freq='2H30min').freq
Out[44]: <150 * Minutes>

gives the correct freq, only not the correct values

[sinhrks]

Ah i see... i haven't understood the phenomenon.

[agraboso]

A question about this has been posted on StackOverflow today.

An interesting observation made there is that

pd.period_range(start='2016-01-01 10:00', freq = '1H1D', periods = 10)

gives the correct (see @sinhrks's comment below) output

PeriodIndex(['2016-01-01 10:00', '2016-01-01 11:00', '2016-01-01 12:00',
             '2016-01-01 13:00', '2016-01-01 14:00', '2016-01-01 15:00',
             '2016-01-01 16:00', '2016-01-01 17:00', '2016-01-01 18:00',
             '2016-01-01 19:00'],
            dtype='int64', freq='25H')

while

pd.period_range(start='2016-01-01 10:00', freq = '1D1H', periods = 10)

(notice the reversal of the freq string) does not.

PeriodIndex(['1971-12-02 01:00', '1971-12-02 02:00', '1971-12-02 03:00',
             '1971-12-02 04:00', '1971-12-02 05:00', '1971-12-02 06:00',
             '1971-12-02 07:00', '1971-12-02 08:00', '1971-12-02 09:00',
             '1971-12-02 10:00'],
            dtype='int64', freq='25H')

[sinhrks]

@agraboso Your example both look incorrect. 1st one should have 25H freq rather than 1H. 1st and 2nd should output the same result.

[agraboso]

@sinhrks You're right, of course. I looked at the first element and the hour on the rest and thought it was fine, but it is not.