should bootstrap algorithms have a fit method? Some, like residual boostrap, invoke forecasters which internally have a fit method. This means - by "inheriting" properties of forecasters - there are computational bottlenecks and index fit vs predict which typically motivate having a fit method.
should bootstrap algorithms allow producing bootstrap samples which are not index identical or length identical with the bootstrapped time series? E.g., for augmentation, benchmarking, or ensembling scenarios. Currently it is assumed produced bootstrap time series are equal length (and implicitly equal index).
Questions from the March 25 meeting:
fit
method? Some, like residual boostrap, invoke forecasters which internally have afit
method. This means - by "inheriting" properties of forecasters - there are computational bottlenecks and index fit vs predict which typically motivate having afit
method.