Currently, the prediction intervals can only be generated using the two provided methods (conformal_distribution & conformal_error). Extending this functionality to custom prediction interval methods would greatly extend the usability of the library.
E.g. by providing a class with a certain signature (e.g. fit method and predict_interval method) that is called in fit and predict and giving the necessary data to these class methods (e.g. the input features, a clone of the model...).
Implementing this inside the library (in contrary to building a wrapper around the MLForecast object) is required in order to include this in the cross-validation pipeline.
Use case
Uncertainty quantification with non built-in methods, when decision making is of high priority.
Description
Currently, the prediction intervals can only be generated using the two provided methods (conformal_distribution & conformal_error). Extending this functionality to custom prediction interval methods would greatly extend the usability of the library. E.g. by providing a class with a certain signature (e.g. fit method and predict_interval method) that is called in fit and predict and giving the necessary data to these class methods (e.g. the input features, a clone of the model...).
Implementing this inside the library (in contrary to building a wrapper around the MLForecast object) is required in order to include this in the cross-validation pipeline.
Use case
Uncertainty quantification with non built-in methods, when decision making is of high priority.