Monoid Est Ton Oid

A post in English and JavaScript.

A while back, I read Hardy Jones’ Comonads, Monoids and Trees (it’s great, if you haven’t read it), and one passage particularly stuck with me:

I have been noticing that whenever there is a reduce around, it is indicative of an abstraction somewhere. Many people call this a code smell.

I’ve been trying to keep this in mind when using reduce, and trying to capture the logic in a monoid. What are monoids, you ask? Well, reader mine, let’s build one and find out…

Motivation

Take this fold (basically just a reduce operation), for example, to find the sum of a list:

const numbers = [1, 2, 3, 4, 5, 6, 7, 8, 9]

// This will produce 45.
numbers.reduce((acc, x) => acc + x, 0)

So, there’s nothing wrong with this code, but what if we need to sum other lists? There are two things that we can reuse:

  • The reduction function, (acc, x) => acc + x
  • The starting value, 0

We’d have a different pair for the product ((acc, x) => acc * x, 1), or for finding the maximum value (Math.max, -Infinity), or anything else. What’s important is that the starting value doesn’t affect the values in the list: for any x, we know it’s always true that 0 + x = x, 1 * x = x, and max(x, -Infinity) = x. We’ll call each of these values the identity for its operation.

Let’s neaten up this concept using a data structure:

const Sum = x => ({
  append: y => Sum(x + y.val),
  val: x
})

Sum.identity = Sum(0)

This structure forms the Sum monoid. We’re there. Bam.

Is That It?

Yep, that’s it. We’re done. There’s honestly no magic to see here. A monoid is a structure that can be appended to other instances of the same structure, and has an identity instance. The only other thing we must do is make sure that the append method is associative:

// For any x, y, and z of the same monoid...
x.append(y).append(z) === x.append(y.append(z))

Just use the intuition that, “as long as the variables are still in the same left-to-right order, the grouping doesn’t matter”. This property is really useful when you come to processing lots of data and you want to split the job up between several threads or nodes: split the data up into chunks, combine (append) all the elements in each chunk, and then combine the chunks. As long as chunk n is appended to the left of chunk n + 1, then everything will Just Work™!

Given this structure, we can now write a tiny function for dealing with it, which we’ll call fold:

const fold = (M, xs) =>
  xs.map(x => M(x)).reduce(
    (acc, x) => acc.append(x),
    M.identity
  )

// Surprise: it's 45!
fold(Sum, numbers).val

We take all the numbers, wrap them in Sum, and then append the Sum instances together to make one overall Sum! Notice that the fold function will also work for other monoids, such as Product and Max. Implement them if you don’t believe me, or you can cheat and jump straight to the JSBin link below!

So, we now only need the fold method for each of our collection structures, and we can reuse these monoids wherever we want. Instead of each structure needing methods for max, length, average, etc, they now only need a fold method, and we can simply pass in the monoid that captures the operation we want to perform. Simpler code, more declarative logic, and more code reuse. Voila.


So, not the longest post, but that’s all from me for today. A lot of functional concepts are much simpler than people would have you believe. If you want to have a play, here are the examples in JSBin. There are many more examples of useful monoids: feel free to research, or try building some of your own.

This post is actually here to provide some background reading for a PHP talk I’m doing next week on monoids and folds. The majority of the talk is around the folding, rather than the monoids, but check out the README for a bigger example of how this all works together.

Take care ♥