The tidy Data Scientist

As a data scientist, having the right tools in your toolbox is a must, and this is one of them. Tidyverse and dplyr verbs allow us to write clean code with the use of tidy evaluation.

# Create a function that takes as arguments a data frame and two column names and returns a filtered data frame
my_filter <- function(df, var1, var2){
 result <- df %>%
 filter({{var1}} > {{var2}})
 return(result)
}

# Create a function that calculate the grouped mean of a variable, passing both as argument
calculate_mean_by_group_var <- function(df, group_var, target_var){
 result <- df %>%
 group_by({{group_var}}) %>% 
 summarise(mean = mean({{target_var}}))
 return(result)
}

Is it possible to use tidy evaluation with ggplot? Say yes!

It just works!! If you are going to create several plots it’s super useful to create wrappers around some of them to avoid repetitive typing.

Let’s plot the closing stock price of TSLA to showcase its use:

# Daily prices from TSLA stock:
head(tsla)
## # A tibble: 6 x 6
## date open high low close volume
## <date> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 2022-01-03 383. 400. 379. 400. 104686047
## 2 2022-01-04 397. 403. 374. 383. 100248258
## 3 2022-01-05 382. 390. 360. 363. 80119797
## 4 2022-01-06 359 363. 340. 355. 90336474
## 5 2022-01-07 360. 360. 337. 342. 84164748
## 6 2022-01-10 333. 353. 327. 353. 91814877
my_plot <- function(df, var){
 df %>%
 ggplot(aes(x=date)) +
 geom_line(aes(y={{var}})) 
}

my_plot(tsla, close)

Easy peasy! Probably you are good to go with that example 90% of the times.

However, another use cases do exist. The main one I can think of is having the column you want to plot stored in another variable. Applying the same solution doesn’t work here.

my_var <- "close"
my_plot(tsla, my_var)

This doesn’t raise an error but it doesn’t produce the desired plot.

Some time ago, we would have solved this with aes_string instead of aes to map the variables, like this:

my_deprecated_plot <- function(df, var){
 df %>%
 ggplot(aes(x=date)) +
 geom_line(aes_string(y=my_var)) 
}

my_var <- "close"
my_deprecated_plot(tsla, my_var)

Whereas it does still work, aes_string has been deprecated so use it at your own risk.

The way we are suppose to overcome this is by using the helper .data inside the mapping call, which I must say it’s pretty neat:

my_plot2 <- function(df, var){
 df %>%
 ggplot(aes(x=date)) +
 geom_line(aes(y=.data[[var]]))
}
my_var <- "close"
my_plot2(tsla, my_var)

There is another corner case that maybe we can face: Having to pass the variable as a character itself, instead of having it in a variable. It’s a silly example as we could simply store it in a variable before calling the plotter but it’s good to know different possibilities.

For this case, again we could simply use the .data helper shown in the previous example, but life would be boring if it was that easy.

Another (weird) way to solve this one is using more advanced and unusual dplyr functions like ensym and the unquote operator from rlang, also known as the bang-bang operator (!!):

my_plot3 <- function(df, var){
 df %>%
 ggplot(aes(x=date)) +
 geom_line(aes(y=!!ensym(var)))
}
my_plot3(tsla, "close")

Definitely the curly-curly operator is really handy when programming with R. I hope you learned something today!

Check out other tidyverse tricks:

• Confidence intervals with ggplot
• Logging pipelines
• Filtering dataframes on multiple conditions
• Conditional pipes