Have you ever...

• written code that produced beautiful output until suddenly it didn't?

• ran a statistical model that - out of the blue - throws a warning?

• copied code from a colleague that just does not work on your computer?

Time-consuming and frustrating issue (typically after updating)

• Popular solution: "If there is one thing I have learnt in R is to never ever update anything!"

quote from and with permission of Lauriane

R, RStudio and R packages

R, RStudio and R packages

Versions to report in a publication

R, RStudio and R packages

The R System library

• Directory1 where packages are stored2 and looked up3 for an analysis

1use .libPaths() to check where, 2stored = install.packages(), 3looked up = library()

• Installation/update of a package == removal of a previous version
  e.g. update to {ggplot2} 3.3.5 == 2.3.9

R package dependencies

• Packages can be linked to other packages (dependent): e.g. {ggplot2} has got 30 dependencies:

Fig.source: RStudio, February 2022

• Installation/update of 1 package == updates all dependencies as well

What does this mean for an analysis?

Example

• Explicit use of 3 packages (by library()):

library(ggplot2) # powerful plotting library
library(lme4) # multi-level modelling
library(naniar) # dealing with missing values

• R implicitly uses 41 packages

Usual analysis

• Explicit use of 10-15 packages

• Implicit interrelations are difficult to gasp & impossible to track by hand

• Issue: (recommended) update unintended consequence

The conceptual solution

The conceptual solution

Isolated project-libraries (also called environments)

The conceptual solution

Isolated project-libraries (also called environments)

Isolated libraries (environments)

• Standard practice in programming languages, e.g. Python

further info: https://environments.rstudio.com/, https://peps.python.org/pep-0405/

• ...one of the main reasons that non-programmers do not proceed with Python

• Fundamentally different to 'usual' interaction with a computer:

  • Imagine the need to define how to use MS Word before creating a new Word-document

• R user = academic researcher, often without any formal training in programming -
  lack of knowledge on situation and solution

The technical framework: {renv}

• R package designed to create reproducible environments
  by initiating isolated project-libraries

Kevin Ushey (2022). renv: Project Environments. R package version 0.15.2.

• What do I see when using {renv}?

  • a subfolder renv that contains the isolated library
    (useful on your computer)

  • a lockfile renv.lock where all versions are documented
    (useful for creating the same library on another computer)

Reflection & key points

• {renv} is not a uniform solution to computational reproducibility
  
  

• Alternatives are available - require more technical know-how
  
  

• {renv} offers a lightweight & manageable solution to

  • setup and use multiple libraries across your computer

  • easily document all versions and dependencies of packages

  • share your environment-setup with other collaborators

Setup {renv}

Official documentation: https://rstudio.github.io/renv/articles/renv.html

1) Open the .Rproj-file of the desired analysis-folder

2) Call install.packages("renv")

3) Call renv::init()

this creates the subfolder and the renv.lock-file

{renv} interacts with you in the console & prints what it does

4) Continue working as usual - if you need to install further packages, they are installed in the project-library

5) From time to time: call renv::snapshot()

adds all the packages used in this project to the renv.lock-file

Setup tested with {renv} 0.15.2, R Version 4.1.3 on Mac OS X 12.2.1

Alternatives

• https://annotater.liomys.mx/ - annotates the version or R packages used in a script (downside: no dependencies, manual installation)

• https://github.com/MilesMcBain/capsule - mimics the lockfile, not the isolated library (downside: need to set-up the environment every time you work on the analysis)

• Setting up everything by hand with {base}-R

Why {renv} ?