Knitr’s best hidden gem: spin

Stop knitting & start spinning – spin can help you write reports much faster and avoid repeating yourself –

Anyone who loves the idea of dynamic report generation with R is probably a big fan of knitr and its flagship function – knit. But not many people seem to know about knit’s awesome cousin – spin.

What is spin?

In short: knitr::spin is similar to knitr::knit in that it takes as input a file with R code + formatting + text and produces a nice readable report, such as an HTML or markdown document. The difference lies in the input: spin operates on R scripts (.R), while knit requires a literate programming file (.Rmd, .Rnw, etc.). This might not sound like a big deal, but it has some important advantages that I’ll describe soon.

Disclaimer: I may have a simplistic and non-comprehensive view of spin and knit; knit certainly has many features that I’ve yet to explore. I’m also going to talk only about the Rmd/md formats and not about any Sweave/LaTeX/HTML for simplicity.

Using spin to turn your R script into markdown output is very simple:

• Any line beginning with #' is treated as a markdown directive (#' # title will be a header, #' some **bold** text results in some bold text)
• Any line beginning with #+ is parsed as code chunk options
• Run knitr::spin on the file

The problem that spin solves

I’ve noticed a problem with the common approach that I adapted (and I know my colleagues use a similar approach) when creating R markdown (.Rmd) documents. There are many usecases to R markdown – whether it be showing the output from a homework assignment, a report to show a peer about your data, or just a report for yourself to more easily see and store results – and a common workflow is this:

• Write a bunch of interactive statements in an .R script
• Once my R code is ready, polish the code and make sure it runs smoothly when sourced
• Copy the code to a .Rmd file, add text with formatting, surround code in code chunks
• Repeat the above several times until the final .Rmd is ready

The main thing to note here is that my whole analysis is now duplicated in two files: in the original R file as regular code, and in the Rmd file as code chunks. I’m breaking the DRY principle (Don’t Repeat Yourself). The problem is that now, because my code is in two places, I have a lot more work to maintain the code. What if next week I want to change a parameter? I’d have to remember to make the exact same change in both the R source code and the Rmd file, so that the report will still match the R script. Having to manage only one source as opposed to keeping two in sync makes my life much simpler.

Enter: spin

With spin, your original R source code is being converted into Rmd and then into a final report format automatically. It’s as if spin is saving you the step of going from R to Rmd. The big benefit here, again, is not just saving the time in writing the Rmd, but the fact that you now have only one copy of your code that you need to maintain.

Once your have some R code that runs, if you want to create a markdown report from it, it’s very easy with spin. You could just go ahead and run spin on pure R, and it’ll work – you’ll get back a markdown file with all the results. It won’t be pretty yet, but it’ll just work right away. To format it better, all you need to do is add some markdown (text) prepended with #', and potentially add chunk options with #+ before every code chunk. You don’t have to copy your code to a different file format and have it be broken until it’s inside a proper code chunk.

Bonus: if you use spin, you can still run your code like a normal R file. You can select a bunch of lines and run them, or just source the entire file, and it’ll run as R code. When you want to create a report, just call spin. If you’re used to writing Rmd files, you may also be a bit annoyed that you can’t simply select lines and run them if those lines are not all inside a code chunk. It’s not an R file, and can’t be treated that way.

Bonus #2: I find it much easier and faster to convert a regular R script to a spin-formatted R script than to a Rmd file. To convert a block of code into a code chunk for spin, you just have to add #+ before the chunk. To convert a block of code into a code chunk in Rmd, you need to add text in two different places – ```{r} before the chunk and ``` after.

People don’t know about spin

The reason I’m writing this is because I believe that spin is very useful but is not very well known.

I was recently at an R workshop taught by Hadley Wickham and chatted with some fellow R enthusiasts about knitr. Out of the 10 useRs I asked, none of them were aware of knitr::spin. I also conducted a similar survey at the labs where I work, and while everyone I asked was an avid user of knitr, nobody knew about spin. I know that my sample size is rather small and I haven’t done any proper statistical significance testing, but I’d say the lack of awareness of spin is quite significant.

knit is still amazing

I’m not trying to say that knit has no place – it’s very popular for a good reason. But in my opinion, before producing a HTML/markdown file from your R code, you should consider whether using spin (pure R code) or knit (need to manually convert your R to Rmd) will be more efficient. I think if the output is very text-heavy and focuses a lot more on formatting and visual, and only includes some bits of R code here and there, then knit would be more appropriate. But if a report is primarily a way to showcase a lot of code, with intermittent text, then spin can save you a lot of trouble.

Example

Ever since my supervisor Jenny Bryan told me about spin, I’ve been using it extensively and it has replaced my use of knit in many circumstances. Every time I have some code that I need to show in a nice format to someone else, I just add some text with #' and add code chunk options where needed.

For example, here is the markdown output from using spin on this R file. It would have been a lot of work to make a Rmd file after writing all the R code, and it would have been annoying to change both files any time in the future.

---

Edit

As has been pointed out on my tweet about this post, you don’t have to call spin in order to get an output from an R file. The “compile notebook” button in RStudio, which calls the rmarkdown::render function, will achieve a similar result if given an R input. This is because render actually calls spin under the hood. It doesn’t really matter if you choose to call spin or to call render – both functions require the same R sript as input rather than an Rmd (or any other literate programming file). The main point here is to show that starting with an R script to produce these outputs is merely possible by using roxygen comments, because I felt that too many people assume their R script has to be manually converted to a different format. So spin/render/notebook away 🙂

There is one nice benefit for render as Jenny Bryan mentioned – it will automatically detect what format your input is, and either call spin or knit or whatever is most appropriate.