Upstream additional resampler functions

[jrnold]

The resampler objects are awesome and exactly what I was looking for in some of my use-cases. As I was using them, I went ahead and implemented several resampling and cross-validation methods. I've been working on them in a package https://github.com/jrnold/resamplr. The additional methods include:

• bootstrap: weighted and Bayesian
• permutations
• jackknife: both ordinary and delete-p
• holdout test/training sets by number or fraction of elements
• k-fold cross validation
• leave-one-out and leave-p-out cross-validation
• time-series k-fold cross-validation

These are all implemented as generic functions with methods for data.frame and grouped_df. Since they are group-aware, all of those methods will work on groups - either treating groups as the element and/or stratifying the method within groups (depending on which are appropriate).

I also have plans and sketched out functions for rolling windows (as in zoo::rollapply), rolling time-series cross-validation (as in Hyndman), time-series bootstrap (as in boot::tsboot), and the remaining options in boot::boot (e.g. balanced bootstrap).

My question is: should I upstream these functions to modelr or continue development as a separate package? I don't want to duplicate effort, and I wanted to ensure they play nicely with modelr/tidyverse.

[jrnold]

I'm going to be doing another rewrite of the resamplr package soon and then submit to CRAN. I'll close this issue when I do that. Let me know if there's anything I can do to make it more compatible with modelr, if you'd rather I upstream those changes, or if you'd like the resampling functions in resamplr to be used in modelr.

[jrnold]

Closing this because rsample package seems to be the logical place for these sort of functions.