Implement STL forecaster

Is your feature request related to a problem? Please describe. STL forecasting seems a common composite forecaster, so it will be useful to provide a class for it.

There is a STLForecaster in statsmodels.

Describe the solution you’d like It should be a composite forecaster consisting of a TransformedTargetPipeline which accepts any other forecaster and applies the forecaster to deseasonalized and detrended data.

forecaster = STLForecaster(MyForecaster(...), ...)

This should be equivalent to this pipeline:

forecaster = TransformedTargetForecaster(
    [
        ("deseasonalise", Deseasonalizer(...)),
        ("detrend", Detrender(forecaster=PolynomialTrendForecaster(degree=1))),
        ("forecast", MyForecaster(...)),
    ]
)

[fkiraly]

Indeed - as a supplementary note, we ran a modified version of test_update_predict_predicted_index_update_params on the instantiated ttf, which takes an object, not a class (i.e., where the “instantiation step” in the test is removed). The test was run with the standard inputs.

[thayeylolu]

I am on it