Similar to how we implemented it for NeuralForecastModel (see here), we should adapt how to pass the model and model_kwargs to StatsForecastModel.
Currently it expects model to be an instance of a statsforecast model. This can be tedious if for example we want to define which models and parameters to use via config files, or that in general we have to import all the models from statsforecast.
Instead, we can leverage the NeuralForecast logic, that allows passing model as a string (or class) and parameters as model_kwargs.
Either we directly enforce this new behavior (a breaking change) or we will deprecate the old behavior in a future release (e.g. deprecation warning for now if model is an instance of a statsforecast model).
Similar to how we implemented it for
NeuralForecastModel(see here), we should adapt how to pass themodelandmodel_kwargstoStatsForecastModel.Currently it expects
modelto be an instance of a statsforecast model. This can be tedious if for example we want to define which models and parameters to use via config files, or that in general we have to import all the models from statsforecast.Instead, we can leverage the NeuralForecast logic, that allows passing
modelas a string (or class) and parameters asmodel_kwargs.Either we directly enforce this new behavior (a breaking change) or we will deprecate the old behavior in a future release (e.g. deprecation warning for now if
modelis an instance of a statsforecast model).