Tips and Tricks for Julia programming

At this point there is a wealth of information for the Julia programming language. Perhaps one of the best sources (outside of the official documentation or discourse forum) is JuliaNotes.jl. Personally, I find that more available information is better and since I'm a big fan of Julia, I've decided to make my own contribution.

Here I'll outline some basic, but at times lesser-known, tips and tricks for writing Julia code.

Ternary Conditional Operator

Julia has the ternary operator for if-else statements. The best way to explain it is to show two equivalent blocks of code. First, one that uses if-else

for i = 1:10
  if i > 5
    println("big")
  else
    println("small")
  end
end

This lengthy if-else statement can be replaced with a ternary operator while still preserving the logic in the code.

for i = 1:10
  i > 5 ? println("big") : println("small")
end

Within this ternary expression, the ? operator denotes which boolean expression (what comes before it) is being evaluated. The : operator separates the if (before the :) and else (after the :) outcomes. This can help shrink lengthy code blocks, but it comes with the additional cost of increased code complexity. However, once familiar with the ternary operator, the complexity of the expression goes away.

Short Circuit Expressions

In Julia the short circuit operators are && (and operator), and || (or operator). The most basic use for them is conditional statements.

x = 5

if (x > 3) && (x < 6)
  println("True")
end

In strictly-typed languages, these expressions can only evaluate to a boolean, as is the case in the example above. However, Julia is a loosely-typed language, which affords it the ability to return the last value in this expression. This becomes a powerful tool for simplifying conditional statements, although some argue that it unnecessarily increases the complexity of the code. Below I show how you can convert the if statement above into a one-line short circuit expression.

x = 5

(x > 3) && (x < 6) && println("True")

This expression is equivalent to the if statement; as such, it only prints True if both conditionals are satisfied (in this case they are). Since both code blocks shown are equivalent, the choice for which to use comes down to personal preference (or the repo style guide). I personally find it convenient to use these statements for continue conditionals within loops.

for i = 1:10
  i == 5 && continue
  println(i)
end

Pipes

If you are familiar with Bash scripting in Linux, then you will already be familiar with pipes. In Julia, the pipe operator is |>. They can be used to redirect the output of something into another function. For example, if you have functions foo and bar such that you can run bar(foo(x)), then you can also rewrite this as foo(x) |> bar. A handy use case for pipes is to avoid defining multiple temporary variables, while also avoiding hard-to-read nested calls.

foo(x) = 2 * sin(x)
bar(x) = 3 * x^2

# Several tmp vars
tmp1 = rand()
tmp2 = foo(tmp1)
x    = bar(tmp2)

# Multiple nested calls
x = bar(foo(rand()))

# One clean easy to read line
x = rand() |> foo |> bar

Anonymous Functions

An anonymous function is (as the name suggests) a function without a name. In Julia, these are done by using the -> operator. A simple f(x) = 2 * x^2 function can be anonymized as x -> 2 * x^2; however, this example does not properly showcase the value of these anonymous functions. Although they can be used in numerous ways, they are most often used in two ways.

1. Alongside pipes

Sometimes you want to pipe output into a function that takes multiple arguments. In those cases, you need to couple a pipe and an anonymous function. The example below shows a simple version of this use case.

foo(x)   = 6 * x^2
bar(x,y) = x^2 - y^2

z = rand() |> foo |> (x -> bar(x, 6))

In this example the output of foo is piped into an anonymous function (x -> bar(x, 6)) which passes the output into bar alongside an additional argument.

2. Within function calls

Many built-in functions take an expression/function as an argument, for instance, filter and find take expressions that evaluate to booleans as arguments.

x = rand(50)

filter(e -> e > 0.5, x)

The code block above uses the anonymous function e -> e > 0.5 as the expression argument for filter, which in this case filters out the values larger than 0.5 from the array x.

Base Functions with Custom Types

This feature involves Julia's multiple dispatch, which is simply when multiple functions have the same name but different arguments. This allows you to write custom base functions for custom types. This is easier to understand through an example.

struct Point{F <: Float64}
  x::F
  y::F
  z::F
end

function Base.show(io::IO, point::Point)
  println(io, "(x,y,z) = ($(point.x),$(point.y),$(point.z))")
end

The example above shows how to make a custom show function for a custom type Point. This is called "extending" the show function. This makes it so that when you display variables of the type Point, they will be displayed using the custom show function above. This means that when instantiating a variable p = Point(1.0, 2.0, 3.0) in the repl, the following output will be (x,y,z) = (1.0,2.0,3.0).

This can be done to all functions you have access to within your Julia code, even those from other imported packages. An important consideration here is to make sure you avoid "type piracy", where you extend a function on types that you did not define. For example,

Base.show(io::IO, f::Float64) = println(io, "This is type piracy")

extending show on the Float64 type (which would not be extending but rather "redefining", since it already exists) is a case of type piracy. This is considered bad practice since it can cause unexpected behavior within your code, or someone else's code if they import your code.