Motivation

Provide some examples of using Tidy evaluation (also called non-standard evaluation (NSE) or delayed evaluation).

ℹ️ This post may be useful for your if you have read Chapters 17-20 of Advanced R book and you are looking to find more additional examples of Tidyeval. You may also want to have a look at the great Tidyeval resource put together by Mara.

Libraries needed for this post

library(rlang)
library(tidyverse)
library(testthat)

Standard function to coefficient of variation

Let’s write a function that calculates coefficient of variation:

cv <- function(var) {
    sd(var) / mean(var)
}

We can test that the function is behaving correctly

testthat::expect_equal( cv(c(3,3)), 0)
testthat::expect_equal( round(cv(c(3,6)),2), 0.47)

Great! The function seems to be doing what we want!

cv <- function(var) {
    var <- enexpr(var) 
    expr(sd(!!var) / mean(!!var)) 
}

testthat::expect_equal(eval(cv(c(3,3))), 0) 
testthat::expect_type((cv(c(3,3))), "language")

Modiying the enviroment - no issues

Let’s modify the function’s environment a little:

cv <- function(var) {
    x=6
    var <- enexpr(var)
    expr(sd(!!var) / mean(!!var))
}

Let’s see what happens:

x=3
testthat::expect_equal( eval(cv(c(3,x))), 0)
x=6
testthat::expect_equal(round(eval(cv(c(3,x))),2), 0.47)

This works even if we put x in the environment where the function is written because x is not an argument of cv function.

Modiying the enviroment - ISSUES!

adding_cv <- function(df,var) {
    x=c(3,6)
    var <- enexpr(var)
    mutate(df, sd(!!var) / mean(!!var))
}

Let’s see what happens:

df<- tibble(n=3)
x=c(3,3)
adding_cv(df,x)

## # A tibble: 1 x 2
##       n `sd(x)/mean(x)`
##   <dbl>           <dbl>
## 1     3           0.471

Wait, what?! The new column contains coefficient of variation. If x=c(3,3) the value in the new column should have been 0. However adding_cv is using x=c(3,6) included in the function environment and not x=c(3,3).

Capturing with enquo

To capture the function and the environment we need enquo

adding_cv <- function(df,var) {
    x=c(3,6)
    var <- enquo(var)
    mutate(df, sd(!!var) / mean(!!var))
}

Let’s see what happens now:

df<- tibble(n=3)
x=c(3,3)
adding_cv(df,x)

## # A tibble: 1 x 2
##       n `sd(x)/mean(x)`
##   <dbl>           <dbl>
## 1     3               0

Changing name with :=

We can make the name of the new column prettier with :=

adding_cv <- function(df,var,nm) {
    x=c(3,3)
    nm_name <- quo_name(nm)
    var <- enquo(var)
    mutate(df, !!nm_name:= sd(!!var) / mean(!!var))
}

Let’s see if that allows changing the name:

x=c(3,6)
df<- tibble(n=3)
adding_cv(df,x,"pretty_name")

## # A tibble: 1 x 2
##       n pretty_name
##   <dbl>       <dbl>
## 1     3       0.471

Insert a list of expressions into a call

What if we want to do delayed filtering? We need to unquote multiple arguments. For this we can use !!!

adding_cv <- function(df,var,nm, ...) {
    x=c(3,6)
    nm_name <- quo_name(nm)
    var <- enquo(var)
    filtering <- enquos(...)
    df %>%  filter(!!!filtering) %>% 
    mutate( !!nm_name:= sd(!!var) / mean(!!var))}

Let’s see if we can filter rows 3 and 6 from column n of our df:

df<- tibble(n=c(3,6,9),m=c(3,6,9))
x=c(3,3)
adding_cv(df,x,"pretty_name", n %in% c(3,6))

## # A tibble: 2 x 3
##       n     m pretty_name
##   <dbl> <dbl>       <dbl>
## 1     3     3           0
## 2     6     6           0

Conclusions

These were just few examples to illustrate why tidyeval can be useful and when it might be needed. I hope it helped you!