Pakillo / Template

template package

Generic template for research or data analysis projects structured as R packages

Rmarkdown documents are great to keep reproducible scientific workflows: tightly integrating code, results and text. Yet, once we are dealing with more complicated data analysis and writing custom code and functions for a project, structuring it as an R package can bring many advantages (e.g. see here and here, or read Marwick et al., but see also here for counterpoints).

Hence this package works as a template for new research or data analysis projects, with the idea of having everything (data, R scripts, functions, and manuscript reporting results) self-contained in the same package (a “research compendium”) to facilitate collaboration and promote reproducibility.

A short presentation introducing this approach on ‘Structuring data analysis projects as R packages’ is available here: https://doi.org/10.6084/m9.figshare.12479984.v1

Installation

# install.packages("remotes")
remotes::install_github("Pakillo/template")

Usage

First, load the package:

library("template")

Now run the function new_project to create a directory with all the scaffolding (slightly modified from R package structure). For example, to start a new project about tree growth, just use:

new_project("treegrowth")

If you want to create a GitHub repository for the project at the same time, use instead:

new_project("treegrowth", github = TRUE, private.repo = FALSE)

You could choose either public or private repository. Note that to create a GitHub repo you will need to have configured your system as explained in https://usethis.r-lib.org/articles/articles/usethis-setup.html.

There are other options you could choose, like setting up testthat or continuous integration (Travis-CI, GitHub actions…). See ?new_project for all options.

Developing the project

1. Now edit README.Rmd and the DESCRIPTION file with some basic information about your project: title, brief description, licence, package dependencies, etc.

2. Place original (raw) data in data-raw folder. Save all R scripts (or Rmarkdown documents) used for data preparation in the same folder.

3. Save final (clean, tidy) datasets in the data folder. You may write documentation for these data.

4. R scripts or Rmarkdown documents used for data analyses may be placed at the analyses folder. The final manuscript/report may be placed at the manuscript folder. You could use one of the many Rmarkdown templates available out there (e.g. rticles, rrtools or rmdTemplates).

5. If you write custom functions, place them in the R folder. Document all your functions with Roxygen. Write tests for your functions and place them in the tests folder.

6. If your analysis uses functions from other CRAN packages, include these as dependencies (Imports) in the DESCRIPTION file (e.g. using usethis::use_package() or rrtools::add_dependencies_to_description(). Also, use Roxygen @import or @importFrom in your function definitions, or alternatively package::function(), to import these dependencies in the namespace.

7. I recommend using an advanced tool like drake or targets to manage your project workflow. A simpler alternative might be writing a makefile or master script to organise and execute all parts of the analysis. A template makefile is included with this package (use makefile = TRUE when calling new_project).

8. Render Rmarkdown reports using rmarkdown::render, and use Rstudio Build menu to create/update documentation, run tests, build package, etc.

9. Record the exact dependencies of your project. One option is simply running sessionInfo() but many more sophisticated alternatives exist. For example, automagic::make_deps_file() or renv::snapshot() will create a file recording the exact versions of all packages used, which can be used to recreate such environment in the future or in another computer. If you want to use Docker, you could use e.g. containerit::dockerfile() or rrtools::use_dockerfile().

10. Archive your repository (e.g. in Zenodo), get a DOI, and include citation information in your README.

Thanks to:

• Carl Boettiger and his template package
• Jeff Hollister and his manuscriptPackage
• Robert Flight: http://rmflight.github.io/posts/2014/07/analyses_as_packages.html
• Hadley Wickham: http://r-pkgs.had.co.nz/
• Yihui Xie: http://yihui.name/knitr/
• Rstudio

Links

• https://github.com/ropensci/rrrpkg
• https://github.com/Reproducible-Science-Curriculum/rr-init
• http://ropensci.github.io/reproducibility-guide/
• https://github.com/jdblischak/r-project-workflows
• https://github.com/benmarwick/rrtools