caret package plans

---

A few people have asked if anything is going to happen to caret now that I’m working at RStudio.

The answer is “very little will be different”.

More specifically:

• caret will be around for a long time. I’m still using it for things too!
• There will still be hills and valleys for development/support. That pattern already exists and will continue since I still have a “day job”.
• As for new things coming out of Rstudio, there is likely to be a fair amount of overlap with caret (in terms of functionality). caret may be the most untidy package out there; just take a look at any call to train or the other modeling functions. There wont be a caret2 but probably a set of packages that overlap with caret in different ways (I thought about trolling and calling the first package carrot).
• caret is fairly focused (myopic?) since it concentrates on prediction. There are similar things that people might want to do where the quantities of interest might not be prediction of a number or class. For example, while you might want to make inferences on your linear mixed model, you might also want to report your resampled RMSE to prove that your model has some fidelity to the actual data.
• There are some features that I want(ed) for caret that would be invasive changes or might break backwards compatibility. Those are more likely to end up in a new package.
• While you might now want a tidyverse solution to modeling (and that’s fine), there will be some features in new pacakges that will be incentives. It is easier to add new stuff in a new system than to redefine the old. Hopefully, I’ve learned my lessons in regards to extensibility; while the user may not realize it, the innards of caret have been refactored more than once.

(This article was originally posted at http://appliedpredictivemodeling.com)