search

unit8co / darts

A python library for user-friendly forecasting and anomaly detection on time series.
https://unit8co.github.io/darts/
Apache License 2.0
8.11k stars 884 forks source link

[BUG] Model page indicating that probabilistic forecasting not applicable for AutoARIMA and StatsforecastAutoCES #2163

Open Jonathan-87 opened 10 months ago

Jonathan-87 commented 10 months ago

In the model page (https://github.com/unit8co/darts?tab=readme-ov-file#forecasting-models) we can see that for AutoARIMA and StatsforecastAutoCES probabilistic forecasting is set to not applicable but pmdarima and statsforecast support prediction intervals for those so I'm wondering why it's indicating that it's not applicable ?

dennisbader commented 10 months ago

It's not really a bug since Darts does not offer support their probabilistic forecasting (yet). The reason why we haven't added this (yet) is to provide a unified probabilistic support across the Darts models.

By default, Darts' probabilistic predictions generate/return samples (from a distribution or simulation). AutoARIMA for example only returns confidence interval bounds.

We could maybe generate multiple confidence intervals, treat them as quantiles and then use the same approach (sampling/direct parmater predictions) as with quantile regression for the regression and torch models. @madtoinou what do you think?