[ENH] AutoETS to ignore multiplicative components when time series is not strictly positive

Is your feature request related to a problem? Please describe. If a time series is not strictly positive, ETS models (from statsmodels) with multiplicative components will raise the following error:

ValueError: endog must be strictly positive when usingmultiplicative error, trend or seasonal components.

Currently AutoETS will raise the same error because it tries all the available ETS models. Ideally AutoETS will skip the models that result in this error and will try all the other ones.

Describe the solution you’d like One solution would be to add a try/except in the _fit method. The other solution I can think of is to check whether the time series is strictly positive and if it’s not remove the multiplicative components from _iter.

The following code reproduces the error:

import pandas as pd
import numpy as np
from sktime.forecasting.ets import AutoETS
from sktime.utils.plotting import plot_series

ts = pd.Series(
    10*np.sin(np.array(range(0,264))/10)+10, 
    pd.date_range("2017-01-01", periods=264,  freq="W")
)
ts[ts<1]=0
forecaster = AutoETS(auto=True, sp=52, n_jobs=-1, ignore_inf_ic=True)
plot_series(ts)
forecaster.fit(ts)

[HYang1996]

Sounds good to me - @HYang1996 would appreciate your opinion if you can find some time?

@mloning, I’m sorry for the late reply. The fable version in R seems to check for this error as well so I think the second approach makes sense.

[mloning]

Thanks @HYang1996! So I suggest we implement that check in our version too and add a short note to the docstring.