Chapter 7 Reproducible Environments
Contributors: Christine Tedijanto and Ben Arnold
7.1 Basics
Running the same code on two different machines is not guaranteed to produce identical results due to potential differences in the computational environment. The computational environment consists of the operating system, installed software, software versions, and more. In this section, we focus on tools for packaging and sharing of computational environments to facilitate reproducibility.
For more on reproducible environments, see the Turing Way section here.
7.2 renv
A package management system is used to track downloaded packages and their versions. One common package management system in R is the renv package. The package has nice documentation that can be found here. When using renv, the state of your project library will be saved to a file in your project called renv.lock.
The most common commands are as follows:
| Command | Description |
|---|---|
renv::init() |
to initialize a project-local environment with a private R library |
renv::snapshot() |
to save the current state of the project library to the renv.lock file |
renv::restore() |
to revert to the state of the project library previously saved in the renv.lock file |
7.3 Docker
Docker is a platform that allows for containerizing environments. The environments can then be shared and launched on a different machine. In our case, we are will often be launching an online browser version of RStudio with a pre-specified R version and packages.
To get started with Docker, download and install Docker on your local machine. Docker is run using the command line, and the accompanying desktop application provides information on which containers and images are currently running. Docker images are the outline for an environment that can be either downloaded or built locally. Images are built using instructions in a text file called a Dockerfile. Although it may take additional time to build an image locally, sharing the Dockerfile allows users to see all the steps that were taken to build the image. A container is an instance of a particular image.
The Rocker Project includes many Docker images that are good starting points for building a reproducible environment with R. More information can be found in this article by the creators of Rocker.
Data can also be shared from your local machine to the Docker container as volumes.
Our experience so far is that Docker can be really great, but it can also be a hassle! The most important step in a reproducible environment is renv (above). Add Docker on top of renv for extra syle points, but never omit renv!
7.4 Putting renv and Docker together
Docker and renv can be combined to define reproducible environments including the operating system, R version, R packages and respective versions, and other dependencies. The renv team has written documentation on how to combine the systems here.
For an example using Docker and renv together to recreate a computational environment, see the repository here.
Other approaches to building reproducible environments include Binder.