etna
etna copied to clipboard
Add `AutoARIMA` model from statsforecast
🚀 Feature Request
As the authors of statsforecast state their auto_arima is 20x faster than the version form pmdarima(which we currently use).There for it might be useful to add adapter for this version too.
Proposal
- Rename current
_AutoARIMAAdapter
to_PMDarimaAutoARIMAAdapter
andPMDarimaAutoARIMAModel
- Create
_StatsForecastAutoARIMAAdapter
for the statsforecast AutoArima version, throw error like here in case of in_sample prediction - Create
StatsForecastAutoARIMAModel(MultiSegmentPredictionIntervalModel)
Test cases
- Check that the
StatsForecastAutoARIMAModel
passes the current tests forPMDarimaAutoARIMAModel
- Add
StatsForecastAutoARIMAModel
to test_inference
Additional context
sktime adapter version link