recalibratiNN package

[cmusso86]

Hi there,

I'm a tidymodels enthusiast interested in post-processing methods. For my bachelor's thesis in statistics, I developed a package implementing a calibration method for Gaussian models, with a special applicability in neural networks optimized by MSE: recalibratiN

During this time, I took the intro course with Max and Hanna at LatinR Uruguay in 2023, and also had the opportunity to talk to them last year at UseR2024 in Salzburg, where I presented the finished package.

During those opportunities, I mentioned this work and asked if it could contribute to the tidymodels universe and they advised me to get in touch. This is my first package, and I'm quite new to this area, so I'm not sure how to format this issue to be considered as a possible contribution... Therefore, I'm looking for some general guidelines on how to proceed. I read the Contribute page on the tidymodels site and browsed the tidyverse style guide.

Should I adapt the package to fit these guidelines and propose a PR directly myself. Or should I just bring this as a discussion topic somehow . Or maybe make this suggestion and, in case you are interested, you guys decide implement something like this in the future?

Any advice would be greatly appreciated.

Thank you so much!

[topepo]

Hey @cmusso86! Good to hear from you.

I'd love to include this. I looked at it in more detail last night and think that it would make a great addition.

It could go into tailor (by adding some wrapper functions) or in probably. The probably package has the other calibration tools in it so making a wrapper there would be the most organized approach. After that, we can invoke that method in tailor in the same way as the current list of methods.

I"ll talk with @simonpcouch too to see if he agrees.

[topepo]

Let's put it in probably. Would you like me to do that?

The machinery in that package is a bit over-engineered. Basically, we would copy what is in functions like cal_estimate_linear() and hook it up to two wrapper functions: one for setting up what is needed for recalibratiNN::recalibrate() and another for running it.

[cmusso86]

That would be great! :)
Thank you!