rrtools: Tools for Writing Reproducible Research in R

Motivation

The goal of rrtools is to provide instructions, templates, and functions for making a basic compendium suitable for writing a reproducible journal article or report with R. This package documents the key steps and provides convenient functions for quickly creating a new research compendium. The approach is based generally on Kitzes et al. (2017), and more specifically on Marwick (2017), Marwick et al. (2018), and Wickham’s (2017) work using the R package structure as the basis for a research compendium.

rrtools provides a template for doing scholarly writing in a literate programming environment using R Markdown and bookdown. It also allows for isolation of your computational environment using Docker, package versioning using MRAN, and continuous integration using Travis. It makes a convenient starting point for writing a journal article or report. If you’re writing a PhD thesis, or a similar type of multi-chapter document, a better choice might the huskydown package or other bookdown variants.

The functions in rrtools allow you to use R to easily follow the best practices outlined in several major scholarly publications on reproducible research. In addition to those cited above, Wilson et al. (2017), Piccolo & Frampton (2016), Stodden & Miguez (2014) and rOpenSci (2017a, b) are important sources that have influenced our approach to this package.

Installation

To explore and test rrtools without installing anything, click the Binder badge above to start RStudio in a browser tab that includes the contents of this GitHub repository. In that environment you can browse the files, install rrtools, and make a test compendium without altering anything on your computer.

You can install rrtools from GitHub with:

# install.packages("devtools")
devtools::install_github("benmarwick/rrtools")

How to use

To create a reproducible research compendium step-by-step using the rrtools approach, follow these detailed instructions. We use RStudio, and recommend it, but is not required for these steps to work. We recommend copy-pasting these directly into your console, and editing the options before running. We don’t recommend saving these lines in a script in your project: they are meant to be once-off setup functions.

0. Create a Git managed directory linked to an online repository

• Usually we want our research compendium to be managed by the version control software Git. The free online book Happy Git With R has details on how to do this. In brief, there are two methods to get started:
  • New project on GitHub first, then download to RStudio: Start on Github, Gitlab, or a similar web service, and create an empty repository called pkgname (you should use a different name, please follow the rules below) on that service. Then clone that repository to have a local empty directory on your computer, called pkgname, that is linked to this remote repository.
  • New project in RStudio first, then connect to GitHub/GitLab: An alternative approach is to create a local, empty, directory called pkgname on your computer, and initialize it with Git (git init), then create a GitHub/GitLab repository and connect your local project to the remote repository.
• Whichever of those two methods that you choose, you continue by staging, commiting an pushing every future change in the repository with Git.
• Your pkgname must follow some rules for everything to work, it must:
  • … contain only ASCII letters, numbers, and ‘.’
  • … have at least two characters
  • … start with a letter (not a number)
  • … not end with ‘.’

1. rrtools::use_compendium("pkgname")

• this uses usethis::create_package() to create a basic R package in the pkgname directory, and then, if you’re using RStudio, opens the project. If you’re not using RStudio, it sets the working directory to the pkgname directory.
• we need to:
  • run rrtools::use_compendium("path/to/pkgname") (you use the path to pkgname in your system)
  • edit the DESCRIPTION file (located in your pkgname directory) to include accurate metadata
  • periodically update the Imports: section of the DESCRIPTION file with the names of packages used in the code we write in the Rmd document(s) (e.g., usethis::use_package("dplyr", "imports"))

2. usethis::use_mit_license(copyright_holder = "My Name")

• this adds a reference to the MIT license in the DESCRIPTION file and generates a LICENSE file listing the name provided as the copyright holder
• to use a different license, replace this line with any of the licenses mentioned here: ?usethis::use_mit_license()

3. rrtools::use_readme_rmd()

• this generates README.Rmd and renders it to README.md, ready to display on GitHub. It contains:
  • a template citation to show others how to cite your project. Edit this to include the correct title and DOI.
  • license information for the text, figures, code and data in your compendium
• this also adds two other markdown files: a code of conduct for users CONDUCT.md, and basic instructions for people who want to contribute to your project CONTRIBUTING.md, including for first-timers to git and GitHub.
• render this document after each change to refresh README.md, which is the file that GitHub displays on the repository home page

4. rrtools::use_analysis()

• this function has three location = options: top_level to create a top-level analysis/ directory, inst to create an inst/ directory (so that all the sub-directories are available after the package is installed), and vignettes to create a vignettes/ directory (and automatically update the DESCRIPTION). The default is a top-level analysis/.
• for each option, the contents of the sub-directories are the same, with the following (using the default analysis/ for example):

analysis/
|
├── paper/
│   ├── paper.Rmd       # this is the main document to edit
│   └── references.bib  # this contains the reference list information

├── figures/            # location of the figures produced by the Rmd
|
├── data/
│   ├── raw_data/       # data obtained from elsewhere
│   └── derived_data/   # data generated during the analysis
|
└── templates
    ├── journal-of-archaeological-science.csl
    |                   # this sets the style of citations & reference list
    ├── template.docx   # used to style the output of the paper.Rmd
    └── template.Rmd

• the paper.Rmd is ready to write in and render with bookdown. It includes:
  • a YAML header that identifies the references.bib file and the supplied csl file (to style the reference list)
  • a colophon that adds some git commit details to the end of the document. This means that the output file (HTML/PDF/Word) is always traceable to a specific state of the code.
• the references.bib file has just one item to demonstrate the format. It is ready to insert more reference details.
• you can replace the supplied csl file with a different citation style from https://github.com/citation-style-language/
• we recommend using the citr addin and Zotero to efficiently insert citations while writing in an Rmd file
• remember that the Imports: field in the DESCRIPTION file must include the names of all packages used in analysis documents (e.g. paper.Rmd). We have a helper function rrtools::add_dependencies_to_description() that will scan the Rmd file, identify libraries used in there, and add them to the DESCRIPTION file.
• this function has an data_in_git = argument, which is TRUE by default. If set to FALSE you will exclude files in the data/ directory from being tracked by git and prevent them from appearing on GitHub. You should set data_in_git = FALSE if your data files are large (>100 mb is the limit for GitHub) or you do not want to make the data files publicly accessible on GitHub.
  • To load your custom code in the paper.Rmd, you have a few options. You can write all your R code in chunks in the Rmd, that’s the simplest method. Or you can write R code in script files in /R, and include devtools::load_all(".") at the top of your paper.Rmd. Or you can write functions in /R and use library(pkgname) at the top of your paper.Rmd, or omit library and preface each function call with pkgname::. Up to you to choose whatever seems most natural to you.

5. rrtools::use_dockerfile()

• this creates a basic Dockerfile using rocker/verse as the base image
• the version of R in your rocker container will match the version used when you run this function (e.g., rocker/verse:3.5.0)
• rocker/verse includes R, the tidyverse, RStudio, pandoc and LaTeX, so compendium build times are very fast on travis
• we need to:
  • edit the Dockerfile to add linux dependencies (for R packages that require additional libraries outside of R). You can find out what these are by browsing the DESCRIPTION files of the other packages you’re using, and looking in the SystemRequirements field for each package. If you are getting build errors on travis, check the logs. Often, the error messages will include the names of missing libraries.
  • modify which Rmd files are rendered when the container is made
  • have a public GitHub repo to use the Dockerfile that this function generates. It is possible to keep the repository private and run a local Docker container with minor modifications to the Dockerfile that this function generates. Or we can use rrtools::use_circleci() to build our Docker container privately at https://circleci.com, from a private GitHub repo.
• If we want to use Travis on our project, we need to make an account at https://hub.docker.com/ to receive our Docker container after a successful build on travis

6. rrtools::use_travis()

• this creates a minimal .travis.yml file. By default it configures travis to build our Docker container from our Dockerfile, and build, install and run our custom package in this container. By specifying docker = FALSE in this function, the travis file will not use Docker in travis, but run R directly on the travis infrastructure. We recommend using Docker because it offers greater computational isolation and saves a substantial amount of time during the travis build because the base image contains many pre-compiled packages.
• we need to:
  • go to https://travis-ci.org/ to connect to our repo
  • add environment variables to enable push of the Docker container to the Docker Hub
  • make an account at https://hub.docker.com/ to host our Docker container
• Note that you should run this function only when we are ready for our GitHub repository to be public. The free travis service we’re using here requires your GitHub repository to be public. It will not work on private repositories. If you want to keep your GitHub repo private until after publication, you can use rrtools::use_circleci() for running free private continuous integration tests at https://circleci.com, instead of travis. With rrtools::use_circleci(docker_hub = FALSE) we can stop our Docker container from appearing on Docker Hub, so our Docker container stays completely private.
• Besides travis and circleci there are also other Continous Integration (CI) services. Gitlab and Github even offer this as part of their repository hosting now. So far rrtools only provides configuration templates for travis and circleci, but we collect examples for other options here.

7. usethis::use_testthat()

• if you add functions in R/, include tests to ensure they function as intended
• create tests.R in tests/testthat/ and check http://r-pkgs.had.co.nz/tests.html for template

You should be able to follow these steps to get a new research compendium repository connected to travis and ready to write in just a few minutes.

References and related reading

Kitzes, J., Turek, D., & Deniz, F. (Eds.). (2017). The Practice of Reproducible Research: Case Studies and Lessons from the Data-Intensive Sciences. Oakland, CA: University of California Press. https://www.practicereproducibleresearch.org

Marwick, B. (2017). Computational reproducibility in archaeological research: Basic principles and a case study of their implementation. Journal of Archaeological Method and Theory, 24(2), 424-450. https://doi.org/10.1007/s10816-015-9272-9

Marwick, B., Boettiger, C., & Mullen, L. (2018). Packaging data analytical work reproducibly using R (and friends). The American Statistician 72(1), 80-88. https://doi.org/10.1080/00031305.2017.1375986

Piccolo, S. R. and M. B. Frampton (2016). “Tools and techniques for computational reproducibility.” GigaScience 5(1): 30. https://gigascience.biomedcentral.com/articles/10.1186/s13742-016-0135-4

rOpenSci community (2017a). Reproducibility in Science A Guide to enhancing reproducibility in scientific results and writing. Online at http://ropensci.github.io/reproducibility-guide/

rOpenSci community (2017b). rrrpkg: Use of an R package to facilitate reproducible research. Online at https://github.com/ropensci/rrrpkg

Schmidt, S.C. and Marwick, B., 2020. Tool-Driven Revolutions in Archaeological Science. Journal of Computer Applications in Archaeology, 3(1), pp.18–32. DOI: http://doi.org/10.5334/jcaa.29

Stodden, V. & Miguez, S., (2014). Best Practices for Computational Science: Software Infrastructure and Environments for Reproducible and Extensible Research. Journal of Open Research Software. 2(1), p.e21. DOI: http://doi.org/10.5334/jors.ay

Wickham, H. (2017) Research compendia. Note prepared for the 2017 rOpenSci Unconf. https://docs.google.com/document/d/1LzZKS44y4OEJa4Azg5reGToNAZL0e0HSUwxamNY7E-Y/edit#

Wilson G, Bryan J, Cranston K, Kitzes J, Nederbragt L, et al. (2017). Good enough practices in scientific computing. PLOS Computational Biology 13(6): e1005510. https://doi.org/10.1371/journal.pcbi.1005510

Contributing

If you would like to contribute to this project, please start by reading uur Guide to Contributing. Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.

Acknowledgements

This project was developed during the 2017 Summer School on Reproducible Research in Landscape Archaeology at the Freie Universität Berlin (17-21 July), funded and jointly organized by Exc264 Topoi, CRC1266, and ISAAKiel. Special thanks to Sophie C. Schmidt for help. The convenience functions in this package are inspired by similar functions in the usethis package.