Tips on best practices in R

This would be a rather standard project structure:

/project_root     /data         /raw
                                      /clean

                      /results      /graphs
                                      /tables

                      /docs
                      /bin
                      /src
                      /ms

You are free to organize your projects in a way that works for you. But following some rather standard way of organizing files will help collaborators find their way around your project and some consistency between projects will also help you adapt scripts and snippets from one project to the next.

If this is how your script starts:

setwd("C:\Users\charlie\some_very_personal_directory_structure\our_project")

what do you think the odds are that the person to whom you are giving that script

• is called Charlie and
  
• is using Windows and
  
• is having the exact same directory structure in their computer?
  

In order to run your script, your friend Lucy, who uses linux, will have to change all links to:

"/home/lucy/totally_different_directory_structure/our_project"

This makes for challenging collaborations.

If you move your project to another machine, the links will equally get broken. And it is likely that they will get broken over time as you reorganize documents in your computer.

Once your entire project lives within one directory (the project root), as we saw in the previous section, make sure that:

This makes your project portable, much more robust (lower risk to get broken over time), and straightforward to share.

The way a lot of people go around this is by setting the working directory manually in RStudio (by clicking on "Set As Working Directory") to the project root. But for this to run in another machine, the user will have to do the same before running the script (unless the script is in the project root, but this is rarely the case and you certainly cannot assume that it is).

A portable script should run as is, without this manual tinkering. Why?

1. It is easy to forget to do the tinkering and wonder why nothing works
   
2. This manual tinkering prevents automation and defeats one of the main advantages of programming:
   

once you realize the advantages of writing code over the use of a GUI software, you should realize that you have to be careful in how you use RStudio. Clicking around is ok for some tasks, but it is not if this does something necessary for the script to run successfully (such as setting the working directory).

If you use RStudio projects (and if you use RStudio, you definitely should create RStudio projects), RStudio will automatically, upon opening a project, move the current working directory to the project root. This is great. But you cannot assume that everybody running your script uses RStudio: R scripts can be run directly in R, in the command line, from a shell script, or using other tools or IDEs. So the script itself should ideally contain a way to refer to the project root, independent of RStudio.

How?

There are various methods, but a wonderfully easy one is:

The function here::here() starts from the current working directory (the directory in which the script is running if you don't set it manually in RStudio or with setwd) and goes up the directory chain until it finds a .Rproj file (if you use RStudio projects), a .git or .svn file (if you version control your projects), a .projectile file (if you use emacs projectile), or other sensible files which signify a project root. If none of these apply to you (which is unlikely), you can create a file .here in your project root with the function set_here() and this file will then signify to the function here() that this is the project root.

From there on, you can refer to any file in your project with here("file/path/from/project/root").

Example usage:

library(tidyverse)
library(here)

my_data <- read_excel(here("data/raw/my_data.xlsx"))
my_plot <- ggplot(data = my_data) + geom_point()
ggsave(here("results/graphs/my_plot.png"))