This post summarises some general and specific points about debugging an rmarkdown file.

General Strategy

In general, users realise they have a problem when something that is working when in manual mode, does not work when they press the knit button in RStudio.

A common cause of this problem is that objects in current workspace are not in the knitr workspace.

it is common to get errors like:

Line ... Error in eval (expr, envir, enclos) : object 'foo' not found

where foo is an object like a data frame or something else you are working with.

In some cases, there will be an obvious error in your code that you can quickly fix up.

After that, a first general step in debugging the problem is to clear the workspace (e.g., by pressing the broom button in rstudio, or typing rm(list=ls()) into the console). Then run each code chunk consecutively starting from the first one in the knitr file. There are shortcut keys in RStudio for running both the current and next code chunk. This can speed up the process. If you get an error, then fix that code.

Sometimes you can get issues, if you issue some commands that are not meant to appear in the knitr documents. In partiuclar, requesting help for a function can cause knitr to hang. So remove any instances of that.

If everything is working after you have cleared the workspace and run all the code manually, its possible that the relationship between the rmarkdown file other files is broken. So, you need to check that the working directory in manual mode is the same as when you run knitr. You can get the current manual-mode working directory by typing getwd(). By default, knitr will use the folder that it is in as the working directory. However, if you are using my modified ProjectTemplate structure, I find it tider to keep the rmarkdown files in a subdirectory called reports. If you do this, then you need to include a command in the first code chunk to tell knitr to treat the working directory as one level up.

Thus, the first code chunk should have the following code inside:

library(knitr) opts_knit$set(root.dir = normalizePath('../')) # required when rmd is in subdirectory

Alternatively, you can just move the rmd file up one level manually and not include the above chunk.

Other assorted errors

Need a later version of rmarkdown: Sometimes RStudio will indicate that it needs a later version of rmarkdown. I've seen in some cases that RStudio's script for this can fail. if so, just install the rmarkdown package with install.packages("rmarkdown")

I often use R markdown documents to organise analyses. In some cases I would like to extract the code and run a subset of named R code chunks. For example, say I had a file foo.rmd:

```{r chunka} print("a") ``` ```{r chunkb} print("b") ``` ```{r chunkc} print("c") ```

I wanted to be able to issue a command that extracts just chunka and chunkb.

The following approach works by creating a new R Markdown document that calls the chunks in the existing document. It is basically based on this answer on StackOverflow.

```{r} library(knitr) purl("foo.rmd") read_chunk("foo.R") ``` ```{r ref.label="chunka"} ``` ```{r ref.label="chunkb"} ```

What is it doing?

purl extracts the R code into a file called foo.R

read_chunk makes the code in foo available for later referencing

I was pulling in a space separated list of variable names from some SPSS syntax into R. I wanted the variable names to be stored as a vector in R.

I used the following to generate the R version of the code:

x <- scan(what=character()) stereotype_assertive stereotype_active stereotype_intelligent stereotype_hardworking stereotype_outgoing stereotype_happy stereotype_attractive stereotype_popular stereotype_successful stereotype_athletic SEXAPP stereotype_feminine stereotype_masculine SEXAPP_male Obesity_Stereotype Glasses_Stereotype Male_Stereotype Halo dput(x)

where all the stuff between scan and the blank line are the variable names. Then dput(x) outputs the character vector

These are just a few notes to self on upgrading R on OSX.

When I upgraded to version 3.1 from 3.0, I couldn't see the old versions of my R packages. While I could have run install.packages(...) where ... is the name of any given package I needed, this would be tedious.

A simpler albeit, slightly riskier approach, was to

go to ~/Library/R and change the folder name from the last installed version number to the updated version of R (e.g., change 3.0 to 3.1)

Open RStudio and update all packages.

References

Location of R packages on OSX

The describe function in the psych package is quite handy. It can be used to obtain descriptive statistics such as mean, sd, min, max, skew, kurtosis, etc. for multiple numeric variables in a data.frame.

However, I recently got an error when I was trying to extract a subset of the summary statistics from the returned object.

Error in x$descript : $ operator is invalid for atomic vectors

The problem seems to be related to the number of classes that the object has. Updating the class to data.frame allows you then treat the returned object as a normal data.frame:

myfoo <- psych::describe(mydata) class(myfoo) <- "data.frame"

So you can then do things like:

myfoo[,c("skew", "kurtosis")]

Often when I google Roxygen or Roxygen2 I have trouble finding documentation. Here's a compilation of some key resources:

Hadley Wickham has a great introduction to core features of Roxygen 2

There are several vignettes listed on the Cran site for Roxygen 2

RStudio has notes on running Roxygen

I have some packages that I am developing and I wanted to get a PDF copy of the documentation for the package.

One way to do this is to first on the command line, make the current directory the one containing the folder with the package source. Then, say the package folder was called foo, run the following.

R CMD Rd2pdf foo/