Month frequency data leads to invalid frequency

[yarnabrina]

Consider this code:

import numpy
import pandas
from sktime.forecasting.naive import NaiveForecaster

numpy.random.seed(0)

times = pandas.to_datetime(["2024-01-01", "2024-02-01", "2024-03-01", "2024-04-01"])
targets = numpy.random.normal(50, 5, 4)

dataset = pandas.Series(targets, index=times)

NaiveForecaster().fit(dataset).predict([1, 2])

This leads to the following error:

ValueError: Invalid frequency. Please select a frequency that can be converted to a regular pd.PeriodIndex. For other frequencies, basic arithmetic operation to compute durations currently do not work reliably.

This works if I change times to day or week frequency:

times = pandas.to_datetime(["2024-01-01", "2024-01-02", "2024-01-03", "2024-01-04"])

or

times = pandas.to_datetime(["2024-01-01", "2024-01-08", "2024-01-15", "2024-01-22"])

This seems related to #582. However, it is possible to actually possible to convert this (month) case to PeriodIndex:

>>> 
>>> times = pandas.to_datetime(["2024-01-01", "2024-02-01", "2024-03-01", "2024-04-01"])
>>> times.to_period()
PeriodIndex(['2024-01', '2024-02', '2024-03', '2024-04'], dtype='period[M]')
>>> 

Question: as a user, how should I modify the inputs to make it work? Since the exact same format works for other frequencies, a consistent format for all frequency/mtype will be very helpful.

[yarnabrina]

@fkiraly created this as discussion as I am not sure if you'll consider this as a pandas bug as the linked PR indicates, or a potential enhancement request in fh.

[fkiraly]

hm, could be considered both a bug or af eature request, as there is not a stated contract for fh in case of irregular time index, and I'm not sure what the expected contract is. It's probably a reasonable expectation that the above code doesn't break...

What would you advise?

[yarnabrina]

I was thinking more towards feature request as it'd support one more type of frequency. But I just discovered that it may also be a potential bug in future because freq attribute of pandas.Timestamp is deprecated, and _coerce_to_period is relying on it.

>>> pandas.to_datetime("2024-01-01").freq
<stdin>:1: FutureWarning: Timestamp.freq is deprecated and will be removed in a future version.