Derive ForecastingHorizon from DatetimeIndex in predict method

Is your feature request related to a problem? Please describe. Basically just a simplification request: converting from DatetimeIndex to ForecastingHorizon seems to me like a step that the BaseForecaster should be able to handle by itself. So instead of:

model = NaiveForecaster()
model.fit(train_data)
fh = ForecastingHorizon(test_data.index, is_relative=False)
forecast_data = model.predict(fh)

I would suggest users should be able to do:

model = NaiveForecaster()
model.fit(train_data)
forecast_data = model.predict(test_data.index)

which would be very elegant imo. It also removes the need to import ForecastingHorizon, which shortens user code and—and here’s my selfish reason—possibly avoids adding sktime as an explicit dependency in libraries that want to be sktime compatible but do not want to depend on it.

Describe the solution you’d like Perhaps the BaseForecaster.predict method can check whether its fh argument is a DatetimeIndex instance and run the necessary conversion itself?

[fkiraly]

I quickly mocked sth up, @Flix6x, in #1169 - it appears the ForecastingHorizon constructor complains when getting a pd.DatetimeIndex instead of doing what I’d think is the obvious thing (from user perspective).

The tests still run for some time and I have to sign off now, but feel free to have a look (and help me debugging in case they break 😃 )

[fkiraly]

Well, thanks for bringing it up, @Flix6x! This behaviour was not on my mental map.