Closed rexxy-sasori closed 4 months ago
Hi @rexxy-sasori,
Bootstrapping is a non-parametric method that does not assume an underlying distribution. Furthermore, it can be applied to any regression model that makes point estimate predictions (no need for probabilistic modeling). Of course, it also has some disadvantages, one of which is the computational cost. I suggest you take a look at this book for more details: https://otexts.com/fpp3/prediction-intervals.html
Thanks. If I am using a probabilistic forecaster, is there even a need to do bootstrapping?
When using probabilistic models, there is no need to use bootstrapping to obtain prediction intervals. However, the current implementation of skforecast is only compatible with regressors that return a single value in its predict
method (point estimate). Therefore, It is likely that the model you are using is not compatible with skforecast.
I know that skforecast uses bootstrapping to estimate prediction uncertainty. How is it different from regressors such as gaussian / bayesian ridge which already gives estimation uncertainty? How can I access the prediction results