datetimeindex.to_period not behaving as expected.

[bmoscon]

Code Sample, a copy-pastable example if possible

import pandas as pd
from datetime import datetime as dt

dates = [
            dt(2000, 1, 1, 0, 0),
            dt(2000, 1, 1, 0, 0, 30),
            dt(2000, 1, 1, 0, 1),
            dt(2000, 1, 1, 0, 1, 30),
            dt(2000, 1, 1, 0, 2),
            dt(2000, 1, 1, 0, 2, 30),
            dt(2000, 1, 1, 0, 3),
            dt(2000, 1, 1, 0, 3, 30),
            dt(2000, 1, 1, 0, 4),
            dt(2000, 1, 1, 0, 4, 30),
            dt(2000, 1, 1, 0, 5),
            dt(2000, 1, 1, 0, 5, 30),
            dt(2000, 1, 1, 0, 6),
            dt(2000, 1, 1, 0, 6, 30)]

idx = pd.DatetimeIndex(dates)
idx.to_period('5T')

Problem description

By changing the period to 5 minutes, I expected the output to be:

PeriodIndex(['2000-01-01 00:00', '2000-01-01 00:00', '2000-01-01 00:00',
             '2000-01-01 00:00', '2000-01-01 00:00', '2000-01-01 00:00',
             '2000-01-01 00:00', '2000-01-01 00:00', '2000-01-01 00:00',
             '2000-01-01 00:00', '2000-01-01 00:05', '2000-01-01 00:05',
             '2000-01-01 00:05', '2000-01-01 00:05'],
            dtype='period[5T]', freq='5T')

but it is

PeriodIndex(['2000-01-01 00:00', '2000-01-01 00:00', '2000-01-01 00:01',
             '2000-01-01 00:01', '2000-01-01 00:02', '2000-01-01 00:02',
             '2000-01-01 00:03', '2000-01-01 00:03', '2000-01-01 00:04',
             '2000-01-01 00:04', '2000-01-01 00:05', '2000-01-01 00:05',
             '2000-01-01 00:06', '2000-01-01 00:06'],
            dtype='period[5T]', freq='5T')

Which makes it look like the actual period is one minute ('T') since thats what the output corresponds to, but the freq shows correctly as 5T.

Have I misunderstood what to_period should return? or is this a bug?

I am using the latest version of pandas (0.23.4)

[TomAugspurger]

cc @jbrockmendel. Not sure if this is expected, but it does look strange.

[jbrockmendel]

That is weird. Possibly related to #17666?

[sinhrks]

Looks the same as #14070.

I think there was an issue regarding Period definition. Current doc says Period is span, but not clear whether pd.Period('2000-01-01 00:01', freq='2T') is invalid or 2T period starting from '2000-01-01 00:01'.

[bmoscon]

I was going by this part of the docs:

http://pandas.pydata.org/pandas-docs/stable/timeseries.html#combining-aliases

the combined offsets '2H20min' look like I'd expect, periods that are spaced 2 hours and 20 minutes apart.

[bmoscon]

as best I can tell the issue is in indexes/period.py

def dt64arr_to_periodarr(data, freq, tz):
    if data.dtype != np.dtype('M8[ns]'):
        raise ValueError('Wrong dtype: %s' % data.dtype)

    freq = Period._maybe_convert_freq(freq)
    base, mult = _gfc(freq)
    return period.dt64arr_to_periodarr(data.view('i8'), base, tz)

_gfc returns the correct frequency code, 8000, but the mult, 5, is not used when constructing the periodarr in _libs/tslibs/period.pyx. The resulting period array thus buckets the original datetimes incorrectly into minute buckets.