ezknitr: R package to avoid the typical working directory pain when using knitr

A new package has hit the CRAN shelves this week. While knitr is one of the most useful R packages in existence, ezknitr is a simple extension to it that adds flexibility in several ways.

One common source of frustration with knitr is that it assumes the directory where the source file lives should be the working directory, which is often not true. ezknitr addresses this problem by giving you complete control over where all the inputs and outputs are, and adds several other convenient features. The two main functions are ezknit() and ezspin(), which are wrappers around knitr’s knit() and spin(), used to make rendering markdown/HTML documents easier.

You can see Jenny Bryan’s way of dealing with this problem in this gist or simply browse the knitr GitHub issues to see people discussing the issue surrounding directories.

Table of contents

• Availability
• Overview
• Motivation & simple use case
• Use case: using one script to analyze multiple datasets
• Experiment with ezknitr
• spin() vs knit()
• Using rmarkdown::render()

Availability

ezknitr is available through both CRAN (install.packages("ezknitr")) and GitHub (devtools::install_github("daattali/ezknitr")).

Overview

If you have a very simple project with a flat directory structure, then knitr works great. But even something as simple as trying to knit a document that reads a file from a different directory or placing the output rendered files in a different folder cannot be easily done with knitr.

ezknitr improves basic knitr functionality in a few ways. You get to decide:

• What the working directory of the source file is
  • Default is your current working directory, which often makes more sense than the knitr assumption that the working directory is wherever the input file is
• Where the output files will go
  • With knitr, all the rendered output files will be generated in the folder you’re currently in
• Where the figures generated in the markdown document will be stored
  • knitr makes it cumbersome to change this directory
• Any parameters to pass to the source file
  • Useful if you want to run an identical source file multiple times with different parameters

Motivation & simple use case

Assume you have an Rmarkdown file that reads a data file and produces a short report while also generating a figure. Native knit() (or spin() if you’re starting with an R script instead of an Rmd file) works great if you have a flat directory structure like this:

- project/
  |- input.csv
  |- report.Rmd

Problem

But what happens if you have a slightly more complex structure? In a real project, you rarely have everything just lying around in the same folder. Here is an example of a more realistic initial directory structure (assume the working directory is set to project/):

- project/
  |- analysis/
     |- report.Rmd
  |- data/
     |- input.csv

Now if you want knitr to work, you’d have to ensure the path to input.csv is relative to the analysis/ directory. This is counter-intuitive because most people expect to create paths relative to the working directory (project/ in this case), but knitr will use the analysis/ folder as the working directory. Any code reading the input file needs to use ../data/input.csv instead of data/input.csv.

Other than being confusing, it also means that you cannot naively run the Rmd code chunks manually because when you run the code in the console, your working directory is not set to what knitr will use as the working directory. More specifically, if you try to run the command that reads the input file, your console will look in project/../data/input.csv (which doesn’t exist).

A similar problem arises when you want to create files in your report: knitr will create the files relative to where the Rmd file is, rather than relative to the project root.

Another problem with the flat directory structure is that you may want to control where the resulting reports get generated. knitr will create all the outputs in your working directory, and as far as I know there is no way to control that.

Solution

ezknitr addresses these issues, and more. It provides wrappers to knit() and spin() that allow you to set the working directory for the input file, and also uses a more sensible default working directory: the current working directory. ezknitr also lets you decide where the output files and output figures will be generated, and uses a better default path for the output files: the directory containing the input file.

Assuming your working directory is currently set to the project/ directory, you could use the following ezknitr command to do what you want:

1 library(ezknitr)
2 ezknit(file = "analysis/report.Rmd", out_dir = "reports", fig_dir = "myfigs")
- project/
  |- analysis/
     |- report.Rmd
  |- data/
     |- input.csv
  |- reports/
     |- myfigs/
        |- fig1.png
     |- report.md
     |- report.HTML

We didn’t explicitly have to set the working direcory, but you can use the wd argument if you do require a different directory. After running ezknit(), you can run open_output_dir() to open the output directory in your file browser if you want to easily see the resulting report. Getting a similar directory structure with knitr is not simple, but with ezknitr it’s trivial.

Note that ezknitr produces both a markdown and an HTML file for each report (you can choose to discard them with the keep_md and keep_html arguments).

Use case: using one script to analyze multiple datasets

As an example of a more complex realistic scenario where ezknitr would be useful, imagine having multiple analysis scripts, with each one needing to be run on multiple datasets. Being the organizer scientist that you are, you want to be able to run each analysis on each dataset, and keep the results neatly organized. I personally was involved in a few projects requiring exactly this, and ezknitr was in fact born for solving this exact issue. Assume you have the following files in your project:

- project/
  |- analysis/
     |- calculate.Rmd
     |- explore.Rmd
  |- data/
     |- human.dat
     |- mouse.dat

We can easily use ezknitr to run any of the analysis Rmarkdowns on any of the datasets and assign the results to a unique output. Let’s assume that each analysis script expects there to be a variable named DATASET_NAME that tells the script what data to operate on. The following ezknitr code illustrates how to achieve the desired output.

1 library(ezknitr)
2 ezknit(file = "analysis/explore.Rmd", out_dir = "reports/human",
3         params = list("DATASET_NAME" = "human.dat"), keep_html = FALSE)
4 ezknit(file = "analysis/explore.Rmd", out_dir = "reports/mouse",
5         params = list("DATASET_NAME" = "mouse.dat"), keep_html = FALSE)
6 ezknit(file = "analysis/calculate.Rmd", out_dir = "reports/mouse",
7         params = list("DATASET_NAME" = "mouse.dat"), keep_html = FALSE)
- project/
  |- analysis/
     |- calculate.Rmd
     |- explore.Rmd
  |- data/
     |- human.dat
     |- mouse.dat
  |- reports/
     |- human/
        |- explore.md
     |- mouse/
        |- calculate.md
        |- explore.md

Note that this example uses the params = list() argument, which lets you pass variables to the input Rmarkdown. In this case, I use it to tell the Rmarkdown what dataset to use, and the Rmarkdown assumes a DATASET_NAME variable exists. This of course means that the analysis script has an external dependency by having a variable that is not defined inside of it. You can use the set_default_params() function inside the Rmarkdown to ensure the variable uses a default value if none was provided.

Also note that differentiating the species in the output could also have been done using the out_suffix argument instead of the out_dir argument. For example, using out_suffix = "human" would have resulted in an ouput file named explore-human.md.

Experiment with ezknitr

After installing the package, you can experiment with ezknitr using the setup_ezknit_test() or setup_ezspin_test() functions to see their benefits. See ?setup_ezknit_test for more information.

spin() vs knit()

knit() is the most popular and well-known function from knitr. It lets you create a markdown document from an Rmarkdown file. I assume if you are on this page, you are familiar with knit() and Rmarkdown, so I won’t explain it any further.

spin() is similar, but starts one step further back: it takes an R script as input, creates an Rmarkdown document from the R script, and then proceeds to create a markdown document from it. spin() can be useful in situations where you develop a large R script and want to be able to produce reports from it directly instead of having to copy chunks into a separate Rmarkdown file. You can read more about why I like spin() in the blog post “knitr’s best hidden gem: spin”.

Using rmarkdown::render()

When the core of this package was developed, none of its functionality was supported in any way by either knitr or rmarkdown. Over time, rmarkdown::render() got some new features that are very similar to features of ezknitr. Native support for parameters inside Rmarkdown files using YAML is a big feature which makes the use of set_default_params() and the params argument of ezknitr less important. However, the core problem that ezknitr wants to solve is the working directory issue, and this issue has yet to be addressed by rmarkdown or knitr, which makes ezknitr still useful.