Instructions for Writing R Tutorials

David Kane

There are no questions here. There are only simple instructions.

Tutorials are not challenging. They are confidence-building.

Every word matters. Never waste a student’s time.

Introduction

This document describes the best way to write R tutorials using the learnr package. The most common use case is a tutorial which covers the material in an assigned textbook, as with the r4ds.tutorials package for R for Data Science (2e) by Hadley Wickham, Mine Çetinkaya-Rundel, and Garrett Grolemund..

Instructors assign textbook readings to students. We want our students to read and, perhaps more importantly, go through the associated code, typing it in and confirming what it does. Sadly, students almost never do so. Fortunately, the tutorial.helpers package provides tools for ensuring that students type in all the assigned code.

Imagine the shallowest possible learning curve. Almost every student should be able to answer almost every Exercise, albeit perhaps with the help of a Hint. There are no hard questions. In fact, there really aren’t any questions at all. Instead, there are instructions: Do one thing, then the next, and then the next.

Assume that you are giving the student a private lesson. You ask them a question. They give you an answer. What would you say to them next? What do you want to teach them, given that context?

There are a million little bits of R (or statistics or data science or . . .) knowledge which we might provide to students: tips, tricks, cool packages, fun websites, et cetera. We don’t have time to mention all of them. The art of teaching is to, first, decide which 10,000 bits are the most important to mention and second, figure out the best time at which to mention them. Tutorials provide an occasion for that mentioning. Which bits do we mention and where do we mention them?

We can only demonstrate a tiny percentage of all the packages in the world, but we can certainly at least mention the existence of dozens more. Within the packages we use, we can only illustrate some of the functions; but we can examine some of the other important ones. Among the functions we use, we can only show some of the arguments; but we can discuss others.

The most important parts of the “knowledge drops” with which we pepper these tutorials are, first, the packages/functions/arguments which we mention without demonstrating, and second, links to high quality resources.

We are building a “pit of success.” Generally, students don’t do the assigned reading, at least in a large class. However, they will complete required work. They will do the assigned tutorials. Our promise: If you complete the tutorials, you will learn the material. There is simply no way not to.

Set Up

Tutorials themselves live in a directory within inst/tutorials in whichever package you are working on. We recommend that this directory name be a combination of a prefix number (which indicate the week/chapter with which a tutorial is associated and/or the order in which to do it) and a name, which corresponds to the id of the tutorial. Within each directory is an R Markdown file and, sometimes, other material like an images or data directory. The prefix number determines the order in which tutorials appear in the Tutorial tab. By default, we name the R Markdown file tutorial.Rmd, but any name works as long as the file has the appropriate YAML header.

To create a new tutorial, use File -> New File -> R Markdown.... Choose the “From Template” option and then select “Helpers Tutorial” from the tutorial.helpers package. Follow the instructions.

The id value is important. It should be the same as the directory in which the tutorial is located, but with any leading numbers removed. It is used for the name of the answer file which students save at the end of the tutorial.

Note that tutorials must be R Markdown documents, meaning that their suffix is .Rmd. You can not (yet) use Quarto documents with tutorials. Fortunately, most of what you need which works in Quarto also works in R Markdown. The main difference is that code chunk options appear within the {}. Don’t worry about this detail. Just use the provided RStudio addin tools.

There is a setup code chunk at the top of a tutorial. You must have library(learnr) and, if you use our tools, library(tutorial.helpers) in this chunk. Our template also provide useful settings for various options.

Warning: You must ensure that any library used in the tutorial is explicitly loaded in this setup chunk. Almost every tutorial makes use of functions from the tidyverse package, so be sure to load this. Unfortunately, nothing in our test suite captures the common error of using library X in the tutorial code and forgetting to load it in the setup chunk.

If your tutorials are part of an R package, then you should ensure that tutorial.helpers is included under Imports and that any library loaded in a tutorial is, at least, included under Suggests.

Structure

The beginning of every tutorial includes the copy-code-chunk and the info-section code chunks. The tutorial is then divided into different Sections that appear as side panels. The first section is the “Introduction” and the last is the “Summary”.

Within the Sections, other than the Introduction and Summary, there are a series of Exercises which can include writing code or writing text. At the end of the tutorial, there is a download-answers code chunk which provides students with instructions on how to download a copy of their answers.

The Introduction section is two to four sentence about the main topics covered in the tutorial. Why are we here? What will students get out of giving you 90 minutes of their lives? What functions/techniques will they learn?

The Summary section is two to four sentences which bring the lessons of the tutorial together for the student. What do they know now that they did not know before? What are the most important functions/techniques covered? This should be very similar to the Introduction. You made a promise as to what they would learn. You kept that promise.

If there are one or two other key resources about the topic of the tutorial, then those resources should be mentioned somewhere in the tutorial and also in the Summary.

Anything typed at the keyboard belongs in `backticks` (not “quotation marks”), except for package names, which are always bolded. Function names always include the parentheses: read_csv(), not read_csv. Example: the + sign is used to connect ggplot() components when using the ggplot2 library.

Ensure that the first few questions always require students to load any libraries which are used in the tutorial. That is, look at all the libraries you load in the set up chunk. (Try not to have too many of them.) All of them, except for learnr and tutorial.helpers merit an exercise which requires the student to type library(package.name). This ensures that students get in the practice of loading libraries. And it also provides occasion to drop some knowledge. Don’t forget that all libraries you load should be included in the DESCRIPTION file — if the tutorial is part of a package — probably under Suggests.

Sections

Tutorials are divided into Sections that appear on the side panel. To create these sections, we include a double hash (##) before the text for it to show up as a side panel. This is also called the Section Title. Use sentence case. On the line after the Section Title, put three hashes. This ensures that students will see the introductory text before they see the first Exercise.

The only section which only has the double hash (##) and not a triple hash (###) on the next line is the Summary section, since there are no Exercises which follow the summary text.

Each Section begins with a sentence or two about what this group of Exercises is trying to accomplish. Example:

## Interacting with sites with `GET()`
###

In order to get data from an API, we use the **httr** package. The package is designed to imitate standard HTTP in R. Read more about HTTP [here](https://www.jmarshall.com/easy/http/).

### Exercise 1

The link will be formatted correctly once the tutorial is knitted. Section introductions will often have two parts: the introductory text as above and a plot which will be replicated in that section of the tutorial. Those two parts are generally separated by a triple hash.

After the last Exercise in a Section, you should put a triple hash and then give a two sentence summary about what this Section accomplished. A Section is a 20 minute transfer of knowledge from you to the student. At the beginning, you mentioned its purpose. Conclude by tying things back to that original purpose. Often, these “purposes” will be fairly trivial: You promised to go through an example of a scatter plot and, in fact, you did. And that is OK! We are not writing poetry. Not every Section leads to salvation.

One or two high quality links, specifically relevant to this Section, should be included/explained at either the beginning or end of a Section, unless the Section is very short.

Exercises

Each Section is composed of a series of numbered Exercises.

Flow

Each Exercise should have a flow which requires that students hit the “Continue” button at least once.

Question types

To create the Exercise headers, you use three hashes. Make sure you number your exercises – ### Exercise 1, ### Exercise 2 and so on.

There are two main types of questions. First, we have normal coding questions. Students write code and press the Run Code button. Second, we have text questions which require students to either write prose or to copy/paste the results of running specific commands. Prose is needed for questions like “Explain the meaning of potential outcomes.” Copy/paste situations arise when students are instructed to do something like connect to Github or edit a qmd. We confirm that the students have completed these questions by having them issue a command like list.files() and then copy/pasting the command and the output. We often abbreviate those instruction using CP/CR, which stands for copy and paste the command and the result.

Code questions

Here is an example code question:

### Exercise 2

Start your code with `cces`. Use the pipe operator `|>` to add the 
function `filter()`, selecting the  rows where `state` is equal to 
"Massachusetts". To set something equal to a value in `filter()` 
use two equal signs: `==`.

```{r filter-2, exercise = TRUE}
```

```{r filter-2-hint-1, eval = FALSE}
cces |> 
  filter(state == "...")
```

```{r filter-2-test, include = FALSE}
cces |> 
  filter(state == "Massachusetts")
```
### 

`==` is used because it is **checking** whether the value of the 
variable on the left is equal to the value on the left. See 
[here](https://stat.ethz.ch/R-manual/R-devel/library/base/html/Comparison.html) 
for discussion of other relational operators in R.
A single equation symbol, `=`, is used to set something equal to 
something else. 

First, the Start of the Exercise sets the stage. It sometimes teaches something new, connects to a previous Exercise, provides a useful link, whatever. If it is long enough, it is followed by a triple hash. If not, the text continues to the instruction. Most of the time, as above, there is only the instruction, telling the student, step-by-step, what to do.

Second, the instruction requires that students write some code. Good instructions generate results when the student presses Run Code. Tutorial answers should require the smallest incremental number of characters, relative to the last question, for students to type. That is one way you know that your learning curve is shallow. If a Exercise code chunk requires the students to type a lot of characters, you should split up the Exercise into multiple separate Exercises.

Third, any Exercise which requires the copying of code from the prior Exercise should place the Copy previous code button below the Exercise code chunk.

<button onclick = "transfer_code(this)">Copy previous code</button>

Fourth, tutorials should be so simple that 95% of the students can answer 95% of the questions easily. One way to ensure that is to add a Hint to almost every coding question.

Hints must always have the same code chunk name as the exercise chunk for which they are the hint, with a “-hint-n” added at the end. So, if an Exercise code chunk is named “ex-1”, then the hint associated with that exercise is named “ex-1-hint-1”. A second hint for that same question would be named “ex-1-hint-2”, and so on.

When you create a hint, always use eval = FALSE within the parentheses in the code chunk. This is because hints will often include “…” and other symbols which do not run as correct R code. So, we need to tell R not to run it or an error will occur during R CMD check. Example:

```{r ex-1-hint-1, eval = FALSE}
This is an example hint. Normally sentences like these 
would cause an error in R because it is not proper code. 
However, since we include eval = FALSE in the r-chunk, 
R ignores all errors!
```

You need to wrap the text in a long hint by hand, inserting the carriage returns yourself. R will not wrap the text automatically.

Often, hints look like this:

```{r ex-1-hint-2, eval = FALSE}
... |> 
  filter(year = ...) |> 
  ...(flights)
```

The “…” indicates places where the student needs to insert some code, a value or a function name. The code in hints should be formatted correctly.

Students can not see the first hint after clicking through to the next hint. So, make sure the last hint is the one you most want them to have access to, i.e., the one which provides the key information. If students can see the last hint, they should have no reason to consult any earlier hints. We rarely provide more that one hint.

Hints are only allowed for coding questions, not for text questions.

Fifth, the third code chunk, after the Exercise and Hint code chunks, is the Test code chunk. Place the answer — the code which you want students to enter into the Exercise code chunk — into the Test code chunk. Since the Test code chunk will be evaluated with the tutorial is knitted (which also happens during testing), this guarantees that correct answer will work for students.

Indeed, the workflow for writing an Exercise often begins by, first, entering the code which we want students to provide into the Test code chunk. We then copy/paste that same code into the Hint code chunk, replacing some of the functions and/or arguments with ... as appropriate. We then ask the question which, we hope, will cause students to answer with the same code as we have in the Test code chunk.

There are some instances in which we can’t test code which we want students to use. The most common case is code which requires the web, generally for downloading data. In that case, we delete the Test code chunk.

Sixth, we separate the code chunks from the end of the Exercise by using ### again.

For simple questions which result in a display of some data, one approach is to write “You should see that the value of height in row 1 is 23.” This allows the students to know that they are on the right track. Never hard-code a number. Use R to inline calculate it, even though this can be a bother.

But, much more common, is to use the end to drop some knowledge, especially about a function which was used in the answer to this Exercise, or to one of the previous Exercises in this Section.

Note that we provide as many Exercises as possible. For example, every tutorial features a question for each package which must be loaded. We require students to type in items like library(tidyverse) even though they have done so many times in the past. Every Exercise is another opportunity to make the learning curve as shallow as possible and to drop some knowledge. More questions are better than fewer.

One way to measure the shallowness of the learning curve is to examine how many new characters each Exercise requires for its answer. Better three Exercises, each of which loads a different package, then one Exercise which loads three packages at once.

Text questions

There are two types of text questions: 1) those that provide the students with the correct answer, after they have submitted their own answer, and, 2) those that do not provide an answer. Examples:

### Exercise 6

Explain potential outcomes in about two sentences.

```{r definitions-6}
question_text(NULL,
    message = "This is where we place the correct answer. It will appear only after 
    students have submitted their own answers. Note that we do not need to wrap the 
    answer text by hand.",          
    answer(NULL, 
           correct = TRUE),
    allow_retry = FALSE,
    incorrect = NULL,
    rows = 6)
``` 

For the message argument, you should provide an excellent answer. We want to allow students to check for themselves that they got, more or less, the correct answer. Note how we set allow_retry to FALSE. This means that, after they see our answer, students can’t modify their answer. The rows argument decides how many rows the empty text input will have.

Always specify (approximately) how much you want students to write. Reasonable units are: one sentence, two sentences and a paragraph. Pick one of these three unless you have a good reason not to.

For paragraph questions, you should mention specific words or phrases which the students should include in their answers. If your suggested answer includes the word “validity,” for example, then tell the students to include (and define) validity as part of their answer.

However, for many written questions, we don’t provide an answer, so we don’t mind if students resubmit. In that case, we use

### Exercise 7

From the Console, run `list.files()`. CP/CR.

```{r file-creation-7}
  question_text(NULL,
    answer(NULL, correct = TRUE),
    allow_retry = TRUE,
    try_again_button = "Edit Answer",
    incorrect = NULL,
    rows = 3)
```

This format is most commonly used for “process” questions in which we have told students to do something and then confirm that they have done it by copying/pasting the result from a command.

tutorial.helpers::show_file() is a handy function for confirming that students have modified text files as instructed. For example, after telling students to edit the _quarto.yml file, we can check that they did so with:

In the Console, run:

```
tutorial.helpers::show_file("_quarto.yml")
```

CP/CR.

show_file() provides a variety of arguments which cause it to return only selected lines rather than the entire file.

Keep in mind that show_file() will not be available to students in their Console by default. We can either always call it with tutorial.helpers::show_file(), as above, or always (and after each restart of the R Session!) have the student type library(tutorial.helpers) at the Console by hand.

Maybe the instructions should always use the double colon, but remind students the first time in each tutorial that they could just library(tutorial.helpers). Or, don’t use the double colon and then, the first time it is used, remind students that, when they see the “No function found” error, they need to library(tutorial.helpers) at the Console.

Tips

Each coding exercise should always spit out something. Interactivity is good! Students should always look at what their code is producing. There are some situations in which students need to make assignments and which, because of this, will result in no output when the Run Code button is pressed. But:

Follow a coding Style Guide, especially spaces around operators like “ = “. Use only one command per line in pipes and graphics, with proper indentation. Indent plotting commands after the call to ggplot().

Do not create an object in one question and then assume that it will be available in subsequent questions. It won’t be! Each question is independent of every other question. They live in separate R instances. The only exception (which we make use of) is that objects created in the initial setup chunk for the entire tutorial are available in all later questions, just the way that library() commands executed there do not need to be executed again.

Tutorials are knitted/run from the directory in which they are located. So, if you want to read in a file from a data/ directory, you write:

x <- read_csv("data/myfile.csv")

in an R code chunk, presumably in the global setup chunk. But, if you try to execute that line of code with Command + Return, it will fail because, by default, you are located in the main directory of your.package when you start working on your tutorials. Using setwd() will solve this problem.

> getwd()
[1] "/Users/davidkane/Desktop/projects/r4ds.tutorials"
> setwd("inst/tutorials/031-data-files/")
> getwd()
[1] "/Users/davidkane/Desktop/projects/r4ds.tutorials/inst/tutorials/031-data-files"
> 

Command + Return will now work because your R session is “located” in the same location as that from which the tutorial will be run when it is knitted.

Pipes

The most common type of code questions involve the step-by-step process of building a pipe, the final output of which is a nice looking graphic.

You want to first show the graphic that you will create by the end of the Section. You show it once at the start of the Section and once before the last Exercise, as a reminder of what the graphic should look like so students do not need to scroll all the way back up.

You should put the code for the graph in the code chunk at the start of the Section. Save the code to an object. The name of the object should have a “_p” suffix, where the “p” stands for “plot.” This way, you only have to put the object name in the code chunk at the end of the Section rather than copying the code.

You then build up the graphic, line by line, over a series of Exercises, providing hints along the way.

Knowledge Drops

The most difficult part of tutorial creation is writing the “knowledge drops,” the snippets of wisdom (and the associated links) which are used at the end of each Exercise. Some advice:

See `?readr::locale` for [details](https://readr.tidyverse.org/articles/locales.html).

Note how we concisely provide both the command which brings up the help page and a link to the help page itself.

Think back to our initial discussion of the 10,000 pieces of information which we want to mention at some point. Among these are many pages linked to from the help pages for the core Tidyverse functions, like arrange() and aes().

Inputs

In addition to tutorial.Rmd, a tutorial will often use other inputs. The two most common locations for storing these inputs are data and images directories at the same level as the tutorial.Rmd file. Any file in data or images will be available at run time. (Note that the directories must have these names. Something like my_data will not work.)

Data

If you need for an R object to be accessible in an Exercise code chunk, create it in the initial global setup code chunk at the top of the tutorial.

Be wary code which downloads data from the web. This won’t work if the student does not have an internet connection when she creates the tutorial. Instead, save the code which downloaded the data and then place that object in an RDS file in the data directory. Here is an example from the “Wrangling Census data with tidyverse tools” tutorial from the tidycensus.tutorials package.

median_age <- get_acs(geography = "county",
                      variables = "B01002_001",
                      year = 2020)
write_rds(median_age, "data/median_age.rds")

median_age <- read_rds("data/median_age.rds")

The first two commands download data and save it to an RDS file in the data directory.

This code assumes that you are located in the same directory as the tutorial.Rmd file. You only run those commands once, and then you comment them out because you don’t want them re-run each time the tutorial is created. The read_rds() call is never commented out because we always need the median_age object.

When designing tutorials which use objects like median_age, we generally write two Exercise code chunks. The first has the student run the same code as that which we used to create the object ourselves. This won’t work if the student is not connected to the web but, with luck, in that case they will get a sensible error message. The second question informs the students that we have, behind the scenes, assigned the result of the function to an R object. We then ask the student to just print out that object. We don’t have them do the assignment themselves, not least because we don’t like questions which don’t generate any output.

We use a similar approach with models which can take awhile to fit. Example:

fit_gauss <- brm(formula = biden ~ 1,
                data = poll_data,
                refresh = 0,
                silent = 2,
                seed = 9)
write_rds(fit_gauss, "data/fit_gauss.rds")

fit_gauss <- read_rds("data/fit_gauss.rds")

Again, this code only works if you are in the tutorial directory, not in the higher directory of the R project itself. Also, the first two commands are commented out, unless you are running them by hand to create the object.

What happens if the data is too large? See the “Arrow” tutorial in the r4ds.tutorials for an example. First, we generally switch away from Code Exercises and use Written Exercises. Students run the required commands and then copy/paste the command/response. Big downloads don’t work well in Exercise code chunks. Second, we create small versions of this big data in the global setup chunk. This allows us to create Test code chunks for most of the exercises which follow. These tests will run much more quickly with this smaller data. Also, for any package on CRAN, we need to keep the overall size of the package as small as possible.

Images

To add images to a tutorial, first make a directory called images in the folder that contains tutorial.Rmd. Store all images for that tutorial there. You can work with those files in all the usual ways.

Use include_graphics() to add the image into the document. Include this code in its own chunk, in the place where you want the image to appear in the tutorial.

```{r}
include_graphics("images/example.png")
```

include_graphics() is part of the knitr package, so you need library(knitr) in the setup code chunk. Note that you do not need to name these code chunks.

Because students will complete the tutorials using screens of very different widths, it is a good idea to put knitr::opts_chunk$set(out.width = '90%') in your setup code chunk. In this way, images will appear as a sensible size regardless of whether they are using a phone screen or a big monitor.

Files

Other files — most commonly text files or RDS files — are handled in a similar fashion to images. Create a directory called data in the same directory as the tutorial.Rmd, just as we do with images.

There are two main uses for files in data. First, they can be used at “compile time” (when the tutorial.Rmd is knit to html) for making plots or doing anything else. Second, and more importantly, they are available to students in the exercise code blocks during “run time” (when students are doing the tutorial).

Complex text

You sometimes want to include “complex” text in a tutorial. This is most common when trying to teach students how to use R code chunks and other strings which Rmarkdown wants to process in certain ways. You can sometimes get away by placing such text in environments surrounded by three, or even four, backticks. This works often, but not always. Comments characters like # are especially problematic. We also use the parsermd package behind the scenes. It does not work as well as one might like.

The “file” trick solves this problem. Create a txt file, example.txt, with the text which you want to appear in the tutorial. You can either leave it in the same directory as the tutorial.Rmd file or, probably better, place it in either the data or images directory. You then add this code chunk to your tutorial.

```{r file = "example.txt", echo = TRUE, eval = FALSE}
```

The file code chunk object loads the specified file. The other code chunk options ensure that the text is echoed but not evaluated.

Processing submissions

Instructors have different needs and priorities when it comes to processing student answers. For now, we provide one function, tutorial.helpers::process_submissions() to help them. Read the help page: ?process_submissions. There are many arguments and options.

Checks

The simplest way to test the tutorial.Rmd with which you are working on is to hit the “Run Document” button. This is the same thing as running rmarkdown::render() on that file:

rmarkdown::render("inst/tutorials/02-terminal/tutorial.Rmd")

This assumes that you are located in the main directory of your.package, as you normally would be. I am not sure if this will catch all potential errors, but it will catch many issues, and it is very quick. Replace 02-terminal with the appropriate directory.

Test before submitting a PR

Once you are done editing your tutorial, you need to make sure it works before you submit a pull request.

  1. Click “Install and Restart” from the Build tab. Then, hit “Start Tutorial” in the Tutorial tab. This mimics the experience that users will have. This will catch some common errors, like having two code chunks with the same name. (I am not sure if this does more or less than the simple test as above.)

  2. Do a full test, which means running R CMD check. Go to the top right window of RStudio. Click the Build pane and hit the “Check” button (with the green check mark). You will then see a bunch of code and tests running. Make sure it says “OK” next to “testthat”. You should always run this before submitting a pull request.

What to do if R CMD check fails

  1. Read the error message at the bottom of the Build pane. You want to see “R CMD check succeeded.” If not, there is a problem. The error message will often provide a clue as to where in your code the error occurred.

  2. If that error message is not detailed enough, go to the your.package.rcheck folder, which should be located in the same directory as your.package on your computer. This is a folder created by the R CMD check process, and it will be automatically deleted if the check process succeeds. If the process fails, the your.package.rcheck folder stays around so that you can examine it. The key file is testthat.Rout.fail, which should be in the tests directory. It has more details on what went wrong.

The most common source of errors is something wrong with the hint code chunks, which are not evaluated when you just Run Document. Make sure the eval = FALSE argument is set in the code chunk for all hints.

Difficult bugs

Error: Cannot find the file(s): "images/rproj.png"

But the file is there! You can see it! The tests work on your local machine. The easiest solution is to delete the file (and commit that change). And then change the name of the file to something else and use it.