One-liners to set-up R projects

Principles

RStudio

All the projects are created with an RStudio project, but you could use starters outside of RStudio.

For RStudio addins fans, @jonmcalder made a nifty RStudio addin for setting up projects with starters.

Best practice: Git, GitHub, Continuous Integration

By default, starters functions encourage the combined use of Git, GitHub and Travis for version control, project sharing, continuous checks and deliveries.

  • We take the stance that you should be using source control and that the source control technology you should be using is Git.

  • For the most part, we expect your Git repositories to live on GitHub. To learn more about git and GitHub we recommend reading Happy Git with R by Jenny Bryan, the STAT 545 TAs, Jim Hester.

  • If you don’t opt out of Git, the projects will be created with a styler pre-commit hook. It’ll run the code styler::style_pkg() to ensure you’re committing properly styled code (you can override it for some commits by using the --no-verify option of git commit).

  • For continuous integration, we only support Travis at the moment, via the tic and travis packages. Travis-CI is a free continuous integration tool that allows projects from GitHub to be checked in a clean environment and for further actions to be taken if the project doesn’t have any errors. This is great for checking out your packages work, but it’s also a really useful utility for analysis and training projects as you can use it to build your reproducible documentation and host it for free on GitHub. If you’re new to Travis, read Julia Silge’s great intro, and create a Travis account linked to your GitHub account.

You can opt out of git by setting the git argument of starters functions to FALSE, and of GitHub as well as Travis by setting the external_setup argument to NULL. See this section for more information about git and external setup options in starters.

When the project is created all the template aspects will be committed to source control, such that you could import the project into GitHub, or use the command line to push it to GitHub.

As all the projects are created with an RStudio project, you should be able to work with the source control aspects within RStudio. You can also use your preferred shell or GUI to interact with it.

Camel or snake?

All project-creating functions in starters have both a snake case name, and a lower camel case name (for historical reasons). E.g. create_basic_project() (snake case) used to be named createBasicProject() (lower camel case). One can use any of these names to call the functions.

Your own adventure

All feedback, bug reports and feature requests, are welcome. Furthermore, if you want to create your own project-creating functions, you could browse starters’ source code for inspiration.

Basic projects and their extensions

Basic projects

A basic project created with create_basic_project() is the fundamental building block for various flavours of projects you might build with starters (analysis projects, training projects).

A basic project contains a minimal package structure to allow you an easy way to:

  • make functions for use within your work,

  • list packages needed for easy dependency management,

  • build your outputs using mechanisms aimed at packages.

Test create_basic_project() via running the chunk below, that’ll create a basic project in a temporary directory, with no dependency management package (to make it run fast), and nothing on GitHub/Travis. The code will also open the new project in a new RStudio session so you can browse files.

All basic projects contain:

├── DESCRIPTION
├── R
├── README.Rmd
├── README.md
└── cooldoggos.Rproj
  • a DESCRIPTION file with default values taken from default DESCRIPTION values you had set (probably at least your name), a Package field equal to your project name and a Title field equal to the title argument. If you let title to its NULL value a random title would be generated using the praise package.

  • an R/ directory

  • a README.Rmd. A README is a wonderful introduction to your project and it’s the first thing people see when they look at it on GitHub. We will create a README.Rmd so that you can write R Markdown and produce a README.md suitable for GitHub. This enables you to show off your functions, your test results, etc. At the top of the README created, there will be a repostatus.org badge indicating the status of your project. By default the initial status is WIP, change it via the initial_status argument. Depending on the external_setup options, a Travis badge might be added too.

  • an RStudio project file

Dependency management

A problem with programming using other people’s code is that other people’s code changes. This risks your results changing over time. To prevent this,

  • starters projects are created with a DESCRIPTION file where you can naturally list dependencies via running usethis::use_package().

  • starters helps you use packages improving dependency management even more, via the packagedeps argument that can be either “none”, “packrat” or “checkpoint” (default).

    • packrat will build a manifest of package versions based on what you use at the point of production. It will then ensure people use the Packrat defined packages, instead of whatever version they have installed. Read more on packrat.

    • checkpoint moves away from using your default CRAN repository and points to a specific repository that has daily snapshots. You then work with CRAN as of a specific date (by default the date you created the project). This is lighter weight than packrat but does not help with non-CRAN packages. Read more on checkpoint.

Git, GitHub, Travis

We take the opinion that your project should feature the following things by default but they can always be turned off by using the parameters in the creation of a project.

  • Git, controlled by the Boolean git argument.

  • external setup, controlled by the external_setup argument, either NULL (no external setup) or a named list. Below is its default value:

Let’s dig into the different elements of the list.

  • “GitHub” is the only supported git_service (Git platform) at the moment. A GitHub repo of the same name as the project will be created. Its description will use the title argument of the create_basic_project() function.

  • The account used will be the one provided as login, if none is provided gh::gh_whoami()$login.

  • The repo created will be public by default (private argument). If you set private to TRUE the repo will be private. It means Travis CI won’t work.

  • You can opt out of “ssh” in favour of “https” via the git_protocol element of the external_setup argument. This will be taken into account whilst setting up the origin remote URL. You’ll still need to write git push --set-upstream origin master for the first push, because of git2r’s current limitations on some operating systems preventing us from setting the upstream.

  • “travis” is the only ci_activation option at the moment, or you can choose NULL for no CI. What will happen is the creation of a Travis config file (.travis.yml), of an empty tic.R file (for other project types, tic.R does contain useful instructions), and the Travis project activation. Since the tic.R file of basic projects is empty, it’s pretty harmless until you add instructions in there.

Analysis projects

Analysis projects are projects where you perform some analysis and probably produce some outputs. The function create_analysis_project() aims to facilitate your analysis by setting up a best practices template project.

At the moment, this function will set up a basic project with some additional folders, by default:

  • data/

  • analysis/

  • outputs/

Test create_analysis_project() via running the chunk below, that’ll create an analysis project in a temporary directory, with no dependency management package (to make it run fast), and nothing on GitHub/Travis. The code will also open the new project in a new RStudio session so you can browse files.

Resulting directory tree:

├── DESCRIPTION
├── R
├── README.Rmd
├── README.md
├── analysis
├── data
├── outputs
└── verycooldoggos.Rproj

Travis

By default, ci_activation of external_setup is equal to “travis”; these projects’ tic.R will knit the R Markdown files contained in analysis/ and deploy them to a gh-pages branch. Note that GitHub pages websites are public.

You can opt out of Travis CI via setting the ci_activation element of the external_setup argument to NULL, e.g.

or even external_setup = NULL to opt out of GitHub as well.

Training projects

Training projects are projects where you need to produce materials for others and/or to produce slides. The function create_training_project() aims to facilitate your training projects by setting up a best practices template project.

At the moment, this function will set up a basic project with some additional folders, by default:

  • data

  • handouts

  • slides

Test create_training_project() via running the chunk below, that’ll create a training project in a temporary directory, with no dependency management package (to make it run fast), and nothing on GitHub/Travis. The code will also open the new project in a new RStudio session so you can browse files.

Resulting directory tree:

├── DESCRIPTION
├── R
├── README.Rmd
├── README.md
├── cat-pictures
├── data
├── handouts
├── slides
└── verynicedoggos.Rproj

Travis

By default, ci_activation of external_setup is equal to “travis”; these projects’ tic.R will knit the R Markdown files contained in handouts/ and slides/ and deploy them to a gh-pages branch. Note that GitHub pages websites are public.

You can opt out of Travis CI via setting the ci_activation element of the external_setup argument to NULL, e.g.

or even external_setup = NULL to opt out of GitHub as well.

Packages used to document

If you make a handouts/ or slides/ directory (both of which are created with the defaults), we will add rmarkdown to the project’s DESCRIPTION file.

You can specify an alternative package for your handouts via handoutEngine. At present, two additional package options are supported. The first is bookdown, and the second is tufte. If you select bookdown or tufte as the handoutEngine, the relevant package will be added to the DESCRIPTION file and the demo files for this handout format will be added to the handouts/ directory.

You can also specify an alternative package for your slides via slideEngine. The currently supported (most popular) package options are revealjs and xaringan. Note also that the default slideEngine option (rmarkdown) caters for a number of additional R Markdown presentation formats such as io_slides, slidy, and beamer since these formats do not have additional package dependencies (although in the case of beamer, you’ll need to have TeX installed). If you are using RStudio, you can find R Markdown templates for these presentation formats by going to File -> New File -> R Markdown -> Presentation.

If the packrat argument is not supplied or is set to TRUE, we will update the packrat manifest so that the packages are available when you open the project.

Package projects

The function create_package_project() aims to facilitate your package development by setting up a best practices template project. All starters projects take advantage of R package building infrastructure, but create_package_project() is the function you’d use as an R package developer to create an R package (or a research compendium).

Test create_package_project() via running the chunk below, that’ll create a package project in a temporary directory, nothing on GitHub/Travis, and will open the new project in a new RStudio session so you can browse files.

Resulting directory tree:

├── CODE_OF_CONDUCT.md
├── DESCRIPTION
├── LICENSE
├── LICENSE.md
├── NAMESPACE
├── NEWS.md
├── R
│   └── coolkittens-package.R
├── README.Rmd
├── README.md
├── _pkgdown.yml
├── coolkittens.Rproj
├── tests
│   ├── testthat
│   │   └── test-sample.R
│   └── testthat.R
└── vignettes
    └── coolkittens.Rmd

Package projects contain

Continuous integration

External setup is controlled like for basic projects with a supplementary argument coverage, either NULL, “codecov” or “coveralls” to control which code coverage, if any, is to be used.

The tic.R of package projects will check the package, build the pkgdown website and deploy it to a gh-pages branch, compute a project health report (see get_project_health()) and deploy it to a project-health branch.

You can opt out of Travis CI via setting the ci_activation element of the external_setup argument to NULL, e.g.

or even external_setup = NULL to opt out of GitHub as well.