9 Functions

Functions should become a really important part of your coding routine. Basically, consider making this your new moto:

Your new moto

If you need to copy-paste, write a function instead.

9.1 Definition

To get the manual on a base function, type ?function_name.

A function returns the last thing that is called in it. Thus a function defined like that will return nothing:

geom_mean <- function(x, y){
    a <- sqrt(x*y)
}
geom_mean(1,2)

While either of these three functions will return what you want:

geom_mean1 <- function(x, y){
    a <- sqrt(x*y)
    a
}
geom_mean2 <- function(x, y){
    a <- sqrt(x*y)
    return(a)
}
geom_mean3 <- function(x, y){
    sqrt(x*y)
}
geom_mean1(1,2) #returns a

#> [1] 1.414214

While an explicit call to return() is not necessary, I think it’s usually a good practise to write it explicitly to make clearer to the reader what the function actually returns.

The result of a function can be a list, a data.frame, a vector… or nothing. And you can attribute the result of a function to a variable:

person <- function(name, age){
    list(name=name, age=age)
}
joe <- person(name="Joe", age=33)
joe

#> $name
#> [1] "Joe"
#> 
#> $age
#> [1] 33

9.2 Default values

One can add default values to variables by specifying them in the function definition:

testfunc <- function(x, y=1){
    x*y
}
testfunc(1)

#> [1] 1

testfunc(1, y=2)

#> [1] 2

testfunc(1:3, y=.1)

#> [1] 0.1 0.2 0.3

9.3 Passing arguments to other functions

One can also pass arguments to other functions within the function, without have to specify all of them:

testfunc2 <- function(x, ...){
    y <- sin(x)
    plot(x, y, ...)
}
# Here, `...` represents all the arguments you could provide to `plot()`
testfunc2(1:100)
testfunc2(1:100, type="l")

You can also pass a function as argument:

testfunc3 <- function(FUN, ...){
    FUN(...)
}
testfunc3(sum, 1:10)

#> [1] 55

testfunc3(plot, 1:10, sin(1:10), type="l")

9.4 Error handling

It is good practise to handle possible errors in your functions, especially if you share them with others. As a general rule of thumb, you can be sure that someone will always be able to find the one case where your function does not work or works in an unexpected way.

For this, the easiest way is to use if() statements together with warning("Warning message") or stop("Stopping message"), to either raise a warning while continuing the function, or stop the function altogether with an explanatory message. Alternatively and for ease of reading, the statement if(condition) stop("Stopping message") can be replaced by stopifnot(condition is true).

Equivalent example with stop() and stopifnot():

test <- function(x, y) {
    if(!is.numeric(x) | !is.numeric(y)){
        stop("Both x and y should numeric vectors, buddy!")
    }
    if(length(x) != length(y)){
        if(length(x) == 1 | length(y) == 1){
            warning(glue::glue("x has length {length(x)}, while y has length {length(y)}, are you sure?"))
        }else{
            stop(glue::glue("Both x and y should have same length or length of 1.\n
            Here x has length {length(x)}, while y has length {length(y)}."))
        }
    }
    return(x*sqrt(y))
}

test(x=1:2,y=1)

#> [1] 1 2

test(x=1:2,y="A")

#> Error in test(x = 1:2, y = "A"): Both x and y should numeric vectors, buddy!

test(x=1:2,y=3:4)

#> [1] 1.732051 4.000000

test(x=1:2,y=3:10)

#> Error in test(x = 1:2, y = 3:10): Both x and y should have same length or length of 1.
#> 
#> Here x has length 2, while y has length 8.

test <- function(x, y) {
    # the condition is the opposite from the one used in stop():
    stopifnot(is.numeric(x), is.numeric(y))
    stopifnot(length(x) == length(y) | length(x) == 1 | length(y) == 1)
    if(length(x) != length(y) & (length(x) == 1 | length(y) == 1)){
        warning(glue::glue("x has length {length(x)}, while y has length {length(y)}, are you sure?"))
    }
    return(x*sqrt(y))
}

test(x=1:2,y=1)

#> [1] 1 2

test(x=1:2,y="A")

#> Error in test(x = 1:2, y = "A"): is.numeric(y) is not TRUE

test(x=1:2,y=3:4)

#> [1] 1.732051 4.000000

test(x=1:2,y=3:10)

#> Error in test(x = 1:2, y = 3:10): length(x) == length(y) | length(x) == 1 | length(y) == 1 is not TRUE

9.5 Documenting your function

It is also good practise to document your function by describing its purpose, its input parameters, its output, and some use cases. For this, the syntax is as follows:

#' Title of the function
#'
#' Description of the function
#' 
#' @param param1 Description of param1
#' @param param2 Description of param2
#'
#' @return Description of the output
#'
#' @examples
#' # Some examples on how to use the function
my_func <- function(param1, param2) {
    # do some stuff
}

RStudio simplifies this process with a tool available in Code > Insert Roxygen skeleton (just place your cursor within the {} of the function before clicking this).

9.6 Interpolation of data

It is possible to interpolate data through the approxfun() and splinefun() functions: while the former uses linear interpolation, the latter uses cubic splines (polynomials).

xmin <- -2*pi; xmax <- 2*pi
x  <- runif(30, min=xmin, max=xmax)
xx <- seq(xmin, xmax, .1)
y  <- sin(x)
# Linear interpolation
lin_interp <- approxfun(x,y)        #is a function
lin_interp(0); lin_interp(pi)

#> [1] -0.00308351

#> [1] -0.000634683

yy_lin     <- approx(x, y, xout=xx) #is a list containing x and y

# Cubic spline interpolation
spl_interp <- splinefun(x,y)        #is a function
yy_spl     <- spline(x, y, xout=xx) #is a list containing x and y

par(family = "Helvetica", cex.lab=1.5, cex.axis=1.4, 
    mgp = c(2.4, .5, 0), tck=0.02, mar=c(4, 4, 2, .5), lwd=2, las=1)
plot(x, y, pch=16, cex=2, ylim=c(-1,1.6))
lines(xx, lin_interp(xx), col="royalblue") # equivalent to: lines(yy_lin, col="royalblue")
lines(xx, spl_interp(xx), col="red", lty=2)# equivalent to: lines(yy_spl, col="red")
legend("topright",
    cex=1.2,
    lty=c(NA, 1, 2),
    lwd=c(NA, 2, 2),
    pch=c(16, NA, NA),
    col=c("black", "royalblue", "red"),
    bty = "n",
    legend=c("'Experimental' points", "Linear interpolation", "Spline interpolation")
    )

9.7 Exercises

Download the archive with all the exercises files, unzip it in your R class RStudio project, and edit the R files.

# Functions Functions should become a really important part of your coding routine. Basically, consider making this your new moto: ::: {.callout-tip} ## Your new moto If you need to copy-paste, write a function instead. ::: ## Definition To get the manual on a base function, type `?function_name`{.R}. A function returns the last thing that is called in it. Thus a function defined like that will return nothing: ```{r, warnings=FALSE} geom_mean <- function(x, y){ a <- sqrt(x*y) } geom_mean(1,2) ``` While either of these three functions will return what you want: ```{r} geom_mean1 <- function(x, y){ a <- sqrt(x*y) a } geom_mean2 <- function(x, y){ a <- sqrt(x*y) return(a) } geom_mean3 <- function(x, y){ sqrt(x*y) } geom_mean1(1,2) #returns a ``` While an explicit call to `return()`{.R} is not necessary, I think it's usually a good practise to write it explicitly to make clearer to the reader what the function actually returns. The result of a function can be a list, a data.frame, a vector... or nothing. And you can attribute the result of a function to a variable: ```{r, warnings=FALSE} person <- function(name, age){ list(name=name, age=age) } joe <- person(name="Joe", age=33) joe ``` ## Default values One can add default values to variables by specifying them in the function definition: ```{r, warnings=FALSE} testfunc <- function(x, y=1){ x*y } testfunc(1) testfunc(1, y=2) testfunc(1:3, y=.1) ``` ## Passing arguments to other functions One can also pass arguments to other functions within the function, without have to specify all of them: ```{r, warnings=FALSE} #| layout-ncol: 2 testfunc2 <- function(x, ...){ y <- sin(x) plot(x, y, ...) } # Here, `...` represents all the arguments you could provide to `plot()` testfunc2(1:100) testfunc2(1:100, type="l") ``` You can also pass a function as argument: ```{r, warnings=FALSE} testfunc3 <- function(FUN, ...){ FUN(...) } testfunc3(sum, 1:10) testfunc3(plot, 1:10, sin(1:10), type="l") ``` ## Error handling It is good practise to handle possible errors in your functions, especially if you share them with others. As a general rule of thumb, you can be sure that *someone* will **always** be able to find the one case where your function does not work or works in an unexpected way. For this, the easiest way is to use `if()`{.R} statements together with `warning("Warning message")`{.R} or `stop("Stopping message")`{.R}, to either raise a warning while continuing the function, or stop the function altogether with an explanatory message. Alternatively and for ease of reading, the statement `if(condition) stop("Stopping message") `{.R} can be replaced by `stopifnot(condition is true)`{.R}. Equivalent example with `stop()`{.R} and `stopifnot()`{.R}: :::: {.columns} ::: {.column width="49%"} ```{r, error=TRUE} test <- function(x, y) { if(!is.numeric(x) | !is.numeric(y)){ stop("Both x and y should numeric vectors, buddy!") } if(length(x) != length(y)){ if(length(x) == 1 | length(y) == 1){ warning(glue::glue("x has length {length(x)}, while y has length {length(y)}, are you sure?")) }else{ stop(glue::glue("Both x and y should have same length or length of 1.\n Here x has length {length(x)}, while y has length {length(y)}.")) } } return(x*sqrt(y)) } ``` ```{r, error=TRUE} test(x=1:2,y=1) test(x=1:2,y="A") test(x=1:2,y=3:4) test(x=1:2,y=3:10) ``` ::: ::: {.column width="2%"} ::: ::: {.column width="49%"} ```{r, error=TRUE} test <- function(x, y) { # the condition is the opposite from the one used in stop(): stopifnot(is.numeric(x), is.numeric(y)) stopifnot(length(x) == length(y) | length(x) == 1 | length(y) == 1) if(length(x) != length(y) & (length(x) == 1 | length(y) == 1)){ warning(glue::glue("x has length {length(x)}, while y has length {length(y)}, are you sure?")) } return(x*sqrt(y)) } ``` ```{r, error=TRUE} test(x=1:2,y=1) test(x=1:2,y="A") test(x=1:2,y=3:4) test(x=1:2,y=3:10) ``` ::: :::: ## Documenting your function It is also good practise to document your function by describing its purpose, its input parameters, its output, and some use cases. For this, the syntax is as follows: ```{r} #' Title of the function #' #' Description of the function #' #' @param param1 Description of param1 #' @param param2 Description of param2 #' #' @return Description of the output #' #' @examples #' # Some examples on how to use the function my_func <- function(param1, param2) { # do some stuff } ``` RStudio simplifies this process with a tool available in *Code > Insert Roxygen skeleton* (just place your cursor within the `{}` of the function before clicking this). ## Interpolation of data It is possible to interpolate data through the `approxfun()`{.R} and `splinefun()`{.R} functions: while the former uses linear interpolation, the latter uses cubic splines (polynomials). ```{r} xmin <- -2*pi; xmax <- 2*pi x <- runif(30, min=xmin, max=xmax) xx <- seq(xmin, xmax, .1) y <- sin(x) # Linear interpolation lin_interp <- approxfun(x,y) #is a function lin_interp(0); lin_interp(pi) yy_lin <- approx(x, y, xout=xx) #is a list containing x and y # Cubic spline interpolation spl_interp <- splinefun(x,y) #is a function yy_spl <- spline(x, y, xout=xx) #is a list containing x and y par(family = "Helvetica", cex.lab=1.5, cex.axis=1.4, mgp = c(2.4, .5, 0), tck=0.02, mar=c(4, 4, 2, .5), lwd=2, las=1) plot(x, y, pch=16, cex=2, ylim=c(-1,1.6)) lines(xx, lin_interp(xx), col="royalblue") # equivalent to: lines(yy_lin, col="royalblue") lines(xx, spl_interp(xx), col="red", lty=2)# equivalent to: lines(yy_spl, col="red") legend("topright", cex=1.2, lty=c(NA, 1, 2), lwd=c(NA, 2, 2), pch=c(16, NA, NA), col=c("black", "royalblue", "red"), bty = "n", legend=c("'Experimental' points", "Linear interpolation", "Spline interpolation") ) ``` ## Exercises {#exo-functions} <a href="Data/exo-in-class.zip" download target="_blank">Download the archive with all the exercises files</a>, unzip it in your `R class` RStudio project, and edit the R files.