Building an R package for the Nutrient Profile Model
Alex Coleman
Read about a recent RSE project that looked at building an R package and some tips along the way.
Building a package in R
Over the last few months I’ve been involved in a research software engineering project working with Vicki Jenneson at the CDRC. This project looked at extending the functionality of their Nutrient Profile Model calculator Shiny app. As part of this I looked at developing a separate R package that implemented a series of functions to represent the Nutrient Profile Model logic and could be applied to product data in bulk.
I’ve used the R language for several years, although predominantly as a scripting language in data science (using the tidyverse), or developing web apps (using Shiny). I haven’t ever written an R package before so this project has been a really interesting experience looking at the R packaging ecosystem.
My main reference point for this experience is the fantastic R packaging book developed by Hadley Wickham and Jennifer Bryan that is available free online and provides a really handy guide for how to go about developing an R package. This naturally led me to the devtools package, which is an R package designed to help you build an R package.
A further tangent on my setup is that I also looked at using a devcontainers for developing this package, which is a personal design choice and not at all necessary but helped ensure a seamless developer experience between different devices!
To get started with writing an R package using devtools is straight forward. Devtools also brings with it the package usethis which includes a number of functions for automatically setting up parts of our package. This includes setting up our initial package folder. We get started with usethis::create_package()
which creates our package folder locally with all the standard boilerplate we need to get started. Usethis is also very useful for integrating additional tools like testthat for testing and roxygen2 for documentation, all through functions that handle initial boilerplate. Adopting devtools, it is worth noting however, will dictate your workflow. The way we’ll write and interact with our developing package will be through a series of devtools functions such as devtools::test()
, and devtools::check()
. I didn’t find any problems with this approach but I can imagine there might be certain flexibility drawbacks if you want to have a super custom setup.
When starting this project I came to it from my previous experience in Python and object orientated programming concepts. Whilst these exist within R it isn’t a truly object oriented language and has a number of different systems for implementing object oriented features. With the time available I briefly dabbled with these but eventually dropped them in favour of adopting a purely functional approach and in particular trying to write something that could behave reasonably nicely with tidyverse packages (although not explicitly depend upon them).
So the basic workflow of developing this package was creating a file within R that contained the functions I wanted for a specific functionality of the package. Writing tests for those functions and saving them in tests/testthat
. Using devtools::test()
to check the tests pass. Writing roxygen2 doc strings on my functions and writing Rmarkdown vignettes showing how to use the package in vignettes
. Putting this all together using devtools::check
to make sure the package built and passed checks for both tests, documentation and vignettes and pushing these to GitHub. I also used pkgdown to build a GitHub pages based documentation website which allowed me to present the internal documentation and vignettes publicly for users. Usethis and pkgdown also come with out of the box GitHub actions setup which is very nice and convenient for setting up CI based documentation deployment.
This all makes for a pretty nice workflow for developing a package. It feels familiar to the Python world where you also typically use a variety of 3rd party packages to help manage your package development (although unlike in Python we aren’t overwhelmed by choice for package build tools). It is however quite a bit of extra complexity, although I recognise tools like devtools are actually trying to reduce this complexity(!), and that is a lot to take on for beginners. This again makes me grateful for the excellent book I’ve used to guide me through this process and would recommend it in a heartbeat for anyone else interested in developing an R package.
Introducing nutrientprofiler
This leads me to introduce the package developed during this project called nutrientprofiler
. This package implements a series of functions that aim to help users assess product nutritional information against the UK Government Nutrient Profile Model. This model was developed by the UK Department of Health to help Ofcom implement restrictions in advertising of foods that are high in fats, salt and sugar (HFSS) and now forms the basis of new legislation restricting in-store and online placement-based promotions.
The CDRC had previously developed a single form Shiny App that implements this model for helping check a product against the Nutrient Profile Model. This package expands on that logic and provides it as explicit low level R functions, along side high level R functions for applying the Nutrient Profile Model across datasets of products. The final part of the RSE project has been integrating the nutrientprofiler
package into the Shiny app to support NPM modelling for multiple products at once.
At present the package is not available via CRAN but can be downloaded directly from GitHub using the remotes
package:
install.packages("remotes")
remotes::install_github("Leeds-CDRC/nutrientprofiler")
Vicki Jenneson, who has helped lead on this project said the following about RSE involvement:
Having support from Alex in the RSE team has been invaluable for this project. Alex was able to unpick a complex challenge and translate it into a simple set of logical steps, providing a well-organised, flexible and transparent data flow within the R package. Alex approached the project with professionalism and took the time to walk me through his work in a way that I could understand, even introducing me to some of the basics of R package building and using Github. I’m excited to have a scalable version of the NPM calculator, thanks to all the work which Alex put in.
Author
Alex Coleman
Research Software Engineer