Learning Julia with #TidyTuesday and Tidier.jl

Learning a new programming language can be tricky, and it can sometimes lead to more ways of getting confused – even in languages you know well! I always mix up len() and length()… When returning to Python after primarily working in R for a while, I found Shiny for Python a really good resource for learning. The concepts were similar enough that writing Shiny code in R and Python didn’t feel too different.

I dabbled in Julia during Advent of Code last year, and wanted to learn a little but more. So when I noticed that Karandeep Singh had started developing Tidier.jl – a Julia implementation of the {tidyverse} – I was very keen to give it a go! So I started #JuliaTuesday – where I used the data sets from #TidyTuesday, but performed the data wrangling and visualisation in Julia instead of R. After 10 weeks, I wanted to share what I’d learned.

What is Julia? < svg class="anchor-symbol" aria-hidden="true" height="26" width="26" viewBox="0 0 22 22" xmlns="http://www.w3.org/2000/svg"> < path d="M0 0h24v24H0z" fill="currentColor"> < path d="M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z">

Let’s back up a second. What is Julia? Julia is an high-level, general-purpose, open source programming language. One of the main differences between Julia and R is that Julia is a compiled language, whereas R is an interpreted language (as is Python). This means it’s likely to be faster than R, and so it piqued my interest. I’m not going to go into the depths of R versus Julia here – plenty of those blog posts already exist! What I want to talk about is Tidier.jl!

Tidier.jl < svg class="anchor-symbol" aria-hidden="true" height="26" width="26" viewBox="0 0 22 22" xmlns="http://www.w3.org/2000/svg"> < path d="M0 0h24v24H0z" fill="currentColor"> < path d="M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z">

Tidier.jl is a Julia implementation of the {tidyverse}. One of its aims is to stick as closely to {tidyverse} syntax as possible – which makes it reasonably straightforward to get started as an R user. In Julia, we can install Tidier.jl using:

1
2

using Pkg
Pkg.add("Tidier")

This is essentially equivalent to using install.packages() in R. Before we dive into a full side-by-side comparison of {tidyverse} and Tidier.jl, let’s load some packages and download the data from the #TidyTuesday GitHub repo. Here, I’m using the US Egg Production data, looking at the number of different types of eggs produced each month in the USA.

1
2
3
4

using Tidier
using UrlDownload
using DataFrames
production = urldownload("https://raw.githubusercontent.com/rfordatascience/tidytuesday/master/data/2023/2023-04-11/egg-production.csv") |> DataFrame ;

1
2
3

library(readr)
library(dplyr)
production <- read_csv('https://raw.githubusercontent.com/rfordatascience/tidytuesday/master/data/2023/2023-04-11/egg-production.csv')

Though the syntax and function names may be quite different between R and Julia here, you can see that the idea is exactly the same: (i) load packages using Using Pkgname or library(Pkgname), and then (ii) read in the data directly from a URL and save it as a dataframe.

Now let’s see how the two compare on some simple data wrangling tasks. Tidier.jl supports a lot of {tidyverse} functions – primarily those originating in {dplyr} and {tidyr}. See tidierorg.github.io/Tidier.jl/stable/ for full documentation on how these functions work in Tidier.jl.

Let’s filter our data to consider only cage-free organic eggs, and then convert the units to millions of eggs:

1
2
3
4

plot_data = @chain production begin
  @filter(prod_process == "cage-free (organic)")
  @mutate(n = n_eggs/1000000)
end

1
2
3

plot_data <- production |>
  filter(prod_process == "cage-free (organic)") |>
  mutate(n = n_eggs/1000000)

The key difference here is chaining in Julia. In practice, add an @ before each function, rather than a |>. The other difference here is that we specify when to begin and end the chain, but otherwise – the similarities are striking.

Data visualisation in Julia < svg class="anchor-symbol" aria-hidden="true" height="26" width="26" viewBox="0 0 22 22" xmlns="http://www.w3.org/2000/svg"> < path d="M0 0h24v24H0z" fill="currentColor"> < path d="M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z">

There are a lot of different plotting packages in Julia, but the one I found myself using most often was AlgebraOfGraphics.jl. AlgebraOfGraphics.jl a a data visualisation language for Julia, that’s built on the idea of combining different building blocks (using + and *) to make plots. The principles are similar to {ggplot2} – where plots are made of layers. There’s even a theme_ggplot2() function included with AlgebraOfGraphics.jl, if you want to stick with the classic theme.

1
2
3
4
5

using AlgebraOfGraphics, CairoMakie
xy = data(plot_data) * mapping(:observed_month, :n) * visual(Lines)
with_theme(theme_ggplot2()) do
draw(xy; axis=(xlabel="", ylabel="Cage-free organic eggs produced (millions)"))
end

1
2
3
4
5

library(ggplot2)
ggplot(data = plot_data,
       mapping = aes(x = observed_month, y = n)) +
  geom_line() +
  labs(x = "", y = "Cage-free organic eggs produced (millions)")

You can see how similar the two plots appear below (the Julia version is on the left!). The only real difference I could spot was that {ggplot2} automatically labels the x-axis with only the year, whereas AlgebraOfGraphics.jl uses the full date.

Another new package has also recently entered the Julia scene – TidierPlots.jl. TidierPlots.jl brings a reimplementation of {ggplot2} to Julia, which is built on top of AlgebraOfGraphics.jl. Although, I didn’t experiment with TidierPlots.jl for my #JuliaTuesday challenge, I couldn’t resist trying it out for this blog post.

At the time of writing, there isn’t yet an implementation of geom_line() in TidierPlots.jl, so we’ll go with points instead:

1
2
3
4
5

using TidierPlots
@ggplot(plot_data, aes(x = observed_month, y = n)) + 
    @geom_point() +
    @labs(x = "",
          y = "Cage-free organic eggs produced (millions)")

It’s so similar to {ggplot2} that you can almost copy and paste your code – and just add an @ at the start of each line! TidierPlots.jl is one of the most exciting developments in Julia, and if you’re familiar with {ggplot2} (or indeed AlgebraOfGraphics.jl), the learning curve is very, very gentle.

Final thoughts < svg class="anchor-symbol" aria-hidden="true" height="26" width="26" viewBox="0 0 22 22" xmlns="http://www.w3.org/2000/svg"> < path d="M0 0h24v24H0z" fill="currentColor"> < path d="M3.9 12c0-1.71 1.39-3.1 3.1-3.1h4V7H7c-2.76.0-5 2.24-5 5s2.24 5 5 5h4v-1.9H7c-1.71.0-3.1-1.39-3.1-3.1zM8 13h8v-2H8v2zm9-6h-4v1.9h4c1.71.0 3.1 1.39 3.1 3.1s-1.39 3.1-3.1 3.1h-4V17h4c2.76.0 5-2.24 5-5s-2.24-5-5-5z">

I really enjoyed getting to grips with Tidier.jl and found it an easier way to start learning Julia, through using concepts and functions that were already familiar to me as an R user. Of course, being proficient in Tidier.jl doesn’t make me proficient in Julia as a whole, but I did get introduced to some of the differences and quirks of Julia along the way. I’m definitely keen to use a little bit more Julia in my work, including with Quarto and in combination with R, through the {JuliaCall} package.

You can view the visualisations I created (with code) in the Quarto document published on QuartoPub. You can also view the source code on GitHub.

Thanks to Karandeep Singh (and other contributors) for developing Tidier.jl – it’s definitely made my Julia journey easier!

Image: giphy.com