Introducing magrittr

Abstract

The magrittr (to be pronounced with a sophisticated french accent) package has two aims: decrease development time and improve readability and maintainability of code. Or even shortr: make your code smokin’ (puff puff)!

To achieve its humble aims, magrittr (remember the accent) provides a new “pipe”-like operator, %>%, with which you may pipe a value forward into an expression or function call; something along the lines of x %>% f, rather than f(x). This is not an unknown feature elsewhere; a prime example is the |> operator used extensively in F# (to say the least) and indeed this – along with Unix pipes – served as a motivation for developing the magrittr package.

This vignette describes the main features of magrittr and demonstrates some features which have been added since the initial release.

Introduction and basics

At first encounter, you may wonder whether an operator such as %>% can really be all that beneficial; but as you may notice, it semantically changes your code in a way that makes it more intuitive to both read and write.

Consider the following example, in which the mtcars dataset shipped with R is munged a little:

library(magrittr)

car_data <- 
  mtcars %>%
  subset(hp > 100) %>%
  aggregate(. ~ cyl, data = ., FUN = . %>% mean %>% round(2)) %>%
  transform(kpl = mpg %>% multiply_by(0.4251)) %>%
  print
#>   cyl   mpg   disp     hp drat   wt  qsec   vs   am gear carb       kpl
#> 1   4 25.90 108.05 111.00 3.94 2.15 17.75 1.00 1.00 4.50 2.00 11.010090
#> 2   6 19.74 183.31 122.29 3.59 3.12 17.98 0.57 0.43 3.86 3.43  8.391474
#> 3   8 15.10 353.10 209.21 3.23 4.00 16.77 0.00 0.14 3.29 3.50  6.419010

We start with a value, here mtcars (a data.frame). From there, we extract a subset, aggregate the information based on the number of cylinders, and then transform the dataset by adding a variable for kilometers per liter as a supplement to miles per gallon. Finally we print the result before assigning it. Note how the code is arranged in the logical order of how you think about the task: data->transform->aggregate, which is also the same order as the code will execute. It’s like a recipe – easy to read, easy to follow!

A horrific alternative would be to write:

car_data <- 
  transform(aggregate(. ~ cyl, 
                      data = subset(mtcars, hp > 100), 
                      FUN = function(x) round(mean(x), 2)), 
            kpl = mpg*0.4251)

There is a lot more clutter with parentheses, and the mental task of deciphering the code is more challenging—particularly if you did not write it yourself.

Note also how “building” a function on the fly for use in aggregate is very simple in magrittr: rather than an actual value as the left-hand side in the pipeline, just use the placeholder. This is also very useful in R’s *apply family of functions.

Granted, you may make the second example better, perhaps throw in a few temporary variables (which is often avoided to some degree when using magrittr), but one often sees cluttered lines like the ones presented.

And here is another selling point: suppose I want to quickly add another step somewhere in the process. This is very easy to do in the pipeline version, but a little more challenging in the “standard” example.

The combined example shows a few neat features of the pipe (which it is not):

  1. By default the left-hand side (LHS) will be piped in as the first argument of the function appearing on the right-hand side (RHS). This is the case in the subset and transform expressions.
  2. %>% may be used in a nested fashion, e.g. it may appear in expressions within arguments. This is illustrated in the mpg to kpl conversion.
  3. When the LHS is needed at a position other than the first, one can use the dot,'.', as placeholder. This is shown in the aggregate expression.
  4. The dot in e.g. a formula is not confused with a placeholder, which is utilized in the aggregate expression.
  5. Whenever only one argument (the LHS) is needed, one can omit the empty parentheses. This is shown in the call to print (which also returns its argument). Here, LHS %>% print(), or even LHS %>% print(.) would also work.
  6. A pipeline with a dot (.) as the LHS will create a unary function. This is used to define the aggregator function.

One feature, which was not demonstrated above is piping into anonymous functions, or lambdas. This is possible using standard function definitions, e.g.:

car_data %>%
(function(x) {
  if (nrow(x) > 2) 
    rbind(head(x, 1), tail(x, 1))
  else x
})

However, magrittr also allows a short-hand notation:

car_data %>%
{ 
  if (nrow(.) > 0)
    rbind(head(., 1), tail(., 1))
  else .
}
#>   cyl  mpg   disp     hp drat   wt  qsec vs   am gear carb      kpl
#> 1   4 25.9 108.05 111.00 3.94 2.15 17.75  1 1.00 4.50  2.0 11.01009
#> 3   8 15.1 353.10 209.21 3.23 4.00 16.77  0 0.14 3.29  3.5  6.41901

Since all right-hand sides are really “body expressions” of unary functions, this is only the natural extension of the simple right-hand side expressions. Of course, longer and more complex functions can be made using this approach.

In the first example, the anonymous function is enclosed in parentheses. Whenever you want to use a function- or call-generating statement as right-hand side, parentheses are used to evaluate the right-hand side before piping takes place.

Another, less useful example is:

1:10 %>% (substitute(f(), list(f = sum)))
#> [1] 55

Additional pipe operators

magrittr also provides three related pipe operators. These are not as common as %>% but they become useful in special cases.

The “tee” pipe, %T>% works like %>%, except it returns the left-hand side value, and not the result of the right-hand side operation. This is useful when a step in a pipeline is used for its side-effect (printing, plotting, logging, etc.). As an example (where the actual plot is omitted here):

rnorm(200) %>%
matrix(ncol = 2) %T>%
plot %>% # plot usually does not return anything. 
colSums
#> [1] -7.060852 -5.397162

The “exposition” pipe, %$% exposes the names within the left-hand side object to the right-hand side expression. Essentially, it is a short-hand for using the with functions (and the same left-hand side objects are accepted). This operator is handy when functions do not themselves have a data argument, as for example lm and aggregate do. Here are a few examples as illustration:

iris %>%
  subset(Sepal.Length > mean(Sepal.Length)) %$%
  cor(Sepal.Length, Sepal.Width)
   
data.frame(z = rnorm(100)) %$% 
  ts.plot(z)

Finally, the “assignment” pipe %<>% can be used as the first pipe in a chain. The effect will be that the result of the pipeline is assigned to the left-hand side object, rather than returning the result as usual. It is essentially shorthand notation for expressions like foo <- foo %>% bar %>% baz, which boils down to foo %<>% bar %>% baz. Another example is:

iris$Sepal.Length %<>% sqrt

The %<>% can be used whenever expr <- ... makes sense, e.g. 

  • x %<>% foo %>% bar
  • x[1:10] %<>% foo %>% bar
  • x$baz %<>% foo %>% bar

Aliases

In addition to the %>%-operator, magrittr provides some aliases for other operators which make operations such as addition or multiplication fit well into the magrittr-syntax. As an example, consider:

rnorm(1000)    %>%
multiply_by(5) %>%
add(5)         %>%
{ 
   cat("Mean:", mean(.), 
       "Variance:", var(.), "\n")
   head(.)
}
#> Mean: 5.185841 Variance: 25.87287
#> [1] 11.802269  4.239720 12.717527  4.076358 -2.113159  4.954990

which could be written in more compact form as:

rnorm(100) %>% `*`(5) %>% `+`(5) %>% 
{
  cat("Mean:", mean(.), "Variance:", var(.),  "\n")
  head(.)
}

To see a list of the aliases, execute e.g. ?multiply_by.

Development

The magrittr package is also available in a development version at the GitHub development page: github.com/tidyverse/magrittr.