Haskell Nuggets: k-means · in Code
AI is hot, so let’s talk about some “classical machine learning” in Haskell with k-means clustering! Let’s throw in some dependent types too.
There are a bazillion ways of implementing such a simple algorithm, but this is how I’d do it, as someone who develops almost exclusively in Haskell (or functional pure languages) in both personal projects and work. It’s not the “right” way or the “best” way, but it’s the way that brings me joy. Hopefully it can also break beyond the simple toy projects you’ll often see in conceptual tutorials. You’ll see how I integrate dependent types, type-driven development, mutable data structures, generating random data, and preparation for parallelism. I have been meaning to shift away from “conceptual” posts and instead post a bit more about small, practical snippets that demonstrate some useful Haskell techniques and principles drive how I approach coding in Haskell overall.
For reference, the intended audience is for people with knowledge of Haskell syntax and basic idioms (mapping, traversing, folding, applicatives). The source code is online here, and is structured as a nix flake script. If you have nix installed (and flakes enabled), you should be able to run the script as an executable (./kmeans.hs). You can also load it for editing with nix develop + ghci.
