Chapter 3 Using functions
3.1 Introduction
Functions are a basic building block of any programming language. To use R effectively—even if our needs are very simple—we need to understand how to use functions. We are not aiming to unpick the inner workings of functions in this course5. The aim of this chapter is to explain what functions are for, how to use them, and how to avoid mistakes when doing so, without worrying too much about how they actually work.
3.2 Functions and arguments
The job of each function in R is to carry out some kind of calculation or computation that would typically require many lines of R code to do “from scratch”. Functions allow us to reuse common computations while offering some control over the precise details of what actually happens. The best way to see what we mean by this is to see one in action. The round function is used to round one or more number(s) to a significant number of digits. To use it, we could type this into the Console and hit Enter:
round(x = 3.141593, digits = 2)We have suppressed the output for now so that we can unpack things a bit first. Every time we use a function we always have to work with the same basic construct (there are a few exceptions, but we can ignore these for now). We start with the name of the function, that is, we use the name of the function as the prefix. In this case, the function name is round. After the function name, we need a pair of opening and closing parentheses. It is this combination of name and parentheses that that alerts R to fact that we are trying to use a function. Whenever we see name followed by opening and closing parentheses we’re seeing a function in action.
What about the bits inside the parentheses? These are called the arguments of the function. That is a horrible name, but it is the one that everyone uses so we have to get used to it. Depending on how it was defined, a function can take zero, one, or more arguments. We will discuss this idea in more detail later in this section.
In the simple example above, we used the round function with two arguments. Each of these was supplied as a name-value pair, separated by a comma. When working with arguments, name-value pairs occur either side of the equals (=) sign, with the name of the argument on the left hand side and the value it takes on the right hand side (notice that the syntax highlighter we used to make this website helpfully colours the argument names differently from the values). The name serves to identify which argument we are working with, and the value is the thing that controls what that argument does in the function.
We refer to the process of associating argument names and values as “supplying the arguments” of the function (sometimes we also say “setting the arguments”). Notice the similarity between supplying function arguments and the assignment operation discussed in the last topic. The difference here is that name-value pairs are associated with the = symbol. This association is also temporary: it only lasts as long as it takes for the function to do whatever it does.
Use = to assign arguments
Do not use the assignment operator <- inside the parentheses when working with functions. This is a “trust us” situation: you will end up in all kinds of difficulty if you do this.
The names of the arguments that we are allowed to use are typically determined for us by the function. That is, we are not free to choose whatever name we like. We say “typically”, because R is a very flexible language and so there are certain exceptions to this simple rule of thumb. For now it is simpler to think of the names as constrained by the particular function we’re using. The arguments control the behaviour of a function. Our job as users is to set the values of these to get the behaviour we want. By now it is now probably fairly obvious what is going to happen when we used the round function like this at the Console:
round(x = 3.141593, digits = 2)## [1] 3.14Remember, we said the round function rounds one or more numbers to a number of significant digits. The argument that specifies the focal number(s) is x; the second argument, digits, specifies the number of decimal places we require. Based on the supplied values of these arguments, 3.141593 and 2, respectively, the round function spits out a value of 3.14, which is then printed to the Console. If we had wanted to the answer to 3 significant digits we would use digits = 3. This is what we mean when we say the values of the supplied arguments controls the behaviour of the function.
3.3 Evaluating arguments and returning results
Whenever R evaluates a function we refer to this action as “calling the function”. In our simple example, we called the round function with arguments x and digits (in this course we treat the phrases “use the function” and “call the function” as synonyms, as the former is more natural to new users). What we have just seen—although it may not be obvious—is that when we call functions they first evaluate their arguments, then perform some kind of action, and finally (optionally) return a value to us when they finish doing whatever it is they do
We will discuss that word “return” in a moment. What do we mean by the word “evaluate” in this context? Take a look at this second example which uses round again:
round(x = 2.3 + 1.4, digits = 0)## [1] 4When we call a function, what typically happens is that everything on the right hand side of an = is first evaluated, the result of this evaluation becomes associated with the corresponding argument name, and then the function does its calculations using the resulting name-value pairs. We say “typically” because other kinds of behaviours are possible—remember, R is a very flexible language—though for the purposes of this course we can assume that what we just wrote is always true. What happened above is that R evaluated 2.3 + 1.4, resulting in the number 3.7, which was then associated with the argument x. We set digits to 0 this time so that round just returns a whole number, 4.
The important thing to realise is that the expression(s) on the right hand side of the = can be anything we like. This third example essentially equivalent to the last one:
myvar <- 2.3 + 1.4
round(x = myvar, digits = 0)## [1] 4This time we created a new variable called myvar and then supplied this as the value of the x argument. When we call the round function like this, the R interpreter spots the fact that something on the right hand side of an = is a variable and associates the value of this variable with x argument. As long as we have actually defined the numeric variable myvar at some point we can use it as the value of an argument.
Keeping in mind what we’ve just learned, take a careful look at this example:
x <- 0
round(x = 3.7, digits = x)## [1] 4What is going on here? The key to understanding this is to realise that the symbol x is used in two different ways here. When it appears on the left hand side of the = it represents an argument name. When it appears on the right hand side it is treated as a variable name, which must have a value associated with it for the above to be valid. This is admittedly a slightly confusing way to use this function, but it is perfectly valid. The message here is that what matters is where things appear relative to the =, not the symbols used to represent them.
We said at the beginning of this section that a function may optionally return a value to us when they finish complete their task. That word “return” is just jargon that refers to the process by which a function outputs a value. If we use a function at the Console this will be the value printed at the end. We can use this value in other ways too. For example, there is nothing to stop us combining function calls with the arithmetic operations:
2 * round(x = 2.64, digits = 0)## [1] 6Here the R interpreter first evaluates the function call, and then multiplies the value it returns by 2. If we want to reuse this value we have to assign the result of function call, for example:
roundnum <- 2 * round(x = 2.64, digits = 0)Using a function with <- is really no different from the examples using multiple arithmetic operations in the last topic. The R interpreter starts on the right hand side of the <-, evaluates the function call there, and only then assigns the value to roundnum.
3.4 Functions do not have “side effects”
There is one more idea about functions and arguments that we really need to understand in order to avoid confusion later on. It relates to how functions modify their arguments, or more accurately, how they do not modify their arguments. Take a look at this example:
myvar <- 3.7
round(x = myvar, digits = 0)## [1] 4myvar## [1] 3.7We created a variable myvar with the value 3.7, rounded this to a whole number with round, and then printed the value of myvar. Notice that the value of myvar has not changed after using it as an argument to round. This is important. R functions typically do not alter the values of their arguments. Again, we say “typically” because there are ways to alter this behaviour if we really want to (yes, R is a very flexible language), but we will never ever do this. The standard behaviour—that functions do not alter their arguments—is what is meant by the phrase “functions do not have side effects”.
If we had meant to round the value of myvar so that we can use this new value later on, we have to assign the result of function evaluation, like this:
myvar <- 3.7
myvar <- round(x = myvar, digits = 0)In this example, we just overwrote the old value, but we could just as easily have created a new variable. The reason this is worth pointing out is that new users sometimes assume certain types of functions will alter their arguments. Specifically, when working with functions manipulate something called a data.frame, there is a tendency to assume that the function changes the data.frame argument. It will not. If we want to make use of the changes, rather than just see them printed to the Console, we need to assign the results. We can do this by creating a new variable or overwriting the old one. We’ll gain first hand experience of this in the Data Wrangling block.
Remember, functions do not have side effects! New R users sometimes forget this and create all kinds of headaches for themselves. Don’t be that person.
3.5 Combining functions
Up until now we have not tried to do anything very complicated in our examples. Using R to actually get useful work almost always involves multiple steps, very often facilitated by a number of different functions. There is more than one way to do this. Here’s a simple example that takes an approach we already know about:
myvar <- sqrt(x = 10)
round(x = myvar, digits = 0)## [1] 3We calculated the square root of the number 10 and assigned the result to myvar, then we rounded this to a whole number and printed the result to the Console. So one way to use a series of functions in sequence is to assign a name to the result at each step and use this as an argument to the function in the next step.
Here is another way to replicate the calculation in the previous example:
round(x = sqrt(x = 10), digits = 0)## [1] 3The technical name for this is function composition. Another way of referring to this kind of expression is as a nested function call: we say that the sqrt function is nested inside the round function. The way to should read these constructs is from the inside out. The sqrt(x = 10) expression is on the right hand side of an = symbol, so this is evaluated first, the result is associated with the x argument of the round function, and only then does the round function do its job.
There aren’t really any new ideas here. We have already seen that the R interpreter evaluates whatever is on the right hand side of the = symbol first before associating the resulting value with the appropriate argument name. However, nested function calls can be confusing at first so we need to see them in action. There’s nothing to stop us using multiple levels of nesting either. Take a look at this example:
round(x = sqrt(x = abs(x = -10)), digits = 0)## [1] 3The abs function takes the absolute value of a number, i.e. removes the - sign if it is there. Remember, read nested calls from the inside out. In this example, first we took the absolute value of -10, then we took the square root of the resulting number (10), and then we rounded this to a whole number.
Nested function calls are useful because they make our R code less verbose (we have to write less), but this comes at the cost of reduced readability. We aim to keep function nesting to a minimum in this book, but we will occasionally have to work with the nesting construct so have to understand it even if we don’t like using it. We’ll see a much-easier-to-read method for applying a series of functions in the Data Wrangling block.
3.6 Specifying function arguments
We have only been working with functions that carry out mathematical calculations with numbers so far. We will see many more in this course as it unfolds. Some functions are designed to extract information about functions for us. For example, take a look at the args function:
args(name = round)## function (x, digits = 0)
## NULLWe can see what args does: it prints a summary of the main arguments of a function to the Console (though it doesn’t always print all the available arguments). What can we learn from the summary of the round arguments? Notice that the first argument, x, is shown without an associated value, whereas the digits part of the summary is printed as digits = 0. The significance of this is that digits has a default value. This means that we can leave out digits when using the round function:
round(x = 3.7)## [1] 4This is obviously the same result as we would get using round(x = 3.7, digits = 0). This is a very useful feature of R, as it allows us keep our R code concise. Some functions take a large number of arguments, many of which are defined with sensible defaults. Unless we need to change these default arguments, we can just ignore them when we call such functions. The x argument of round does not have a default, which means we have to supply a value. This is sensible, as the whole purpose of round is to round any number we give it.
There is another way to simplify our use of functions. Take a look at this example:
round(3.72, digits = 1)## [1] 3.7What does this demonstrate? We do not have to specify argument names, i.e. there is no need to specify argument names. In the absence of an argument name the R interpreter uses the position of the supplied argument to work out which name to associate it with. In this example we left out the name of the argument at position 1. This is where x belongs, so we end up rounding 3.71 to 1 decimal place. R is even more flexible than this, as it carries out partial matching on argument names:
round(3.72, dig = 1)## [1] 3.7This also works because R can unambiguously match the argument I named dig to digits. Take note, if there were another argument to round that started with the letters dig this would have caused an error. We have to know our function arguments if we want to rely of partial matching.
Be careful with your arguments
Here is some advice. Do not rely on partial matching of function names. It just leads to confusion and the odd error. If you use it a lot you end up forgetting the true name of arguments, and if you abbreviate too much you create name matching conflicts. For example, if a function has arguments arg1 and arg2 and you use the partial name a for an argument, there is no way to know which argument you meant. We are pointing out partial matching so that you are aware of the behaviour. It is not worth the hassle of getting it wrong just to save on a little typing, so do not use it.
What about position matching? This can also cause problems if we’re not paying attention. For example, if you forget the order of the arguments to a function and then place your arguments in the wrong place, you will either generate an error or produce a nonsensical result. It is nice not to have to type out the name = value construct all the time though, so our advice is to rely positional matching only for the first argument. This is a common convention in R, and it makes sense because it is often obvious what kind of information or data the first argument should carry, so the its name is redundant.
At some point, if you really want to understand what happens when you use a function you will need to grapple with ideas like lazy evaluation, environments and scoping.↩