Linear regression in Rust

Wed, 20 May 2026 10:03:01 +0200

[2026-07-17]
This is a short note I wrote two months ago when I was studying the design of the [ndarray][1] Rust library. I intended to complete the project with some tests and output some plot using any Gnuplot crate I would find but I lost interest in this project in the mean time. I guess this is just for posterity now. However, it was a fun exercise and I really enjoyed malaxing some Rust code for a statistical toy project.

Lately, I’ve been playing with some Rust. I found a toy example of linear regression in Rust: Linear Regression in Rust. Be sure to check the associated GitHub repository as the sample code from the blog post isn’t expanded to its full length. This really was a quick check to see what’s available in Rust, and how developing in Rust feels when using Neovim. About the second point, I want to be able to run tests directly from within my editor, as well as launch a debugger without relying on external task file or firing up another terminal. Note that I removed uninteresting part of the original code (linear regression implemented using standard formulae based on summated standard deviations).

I got several error with cargo run due to openblas-build requiring the rustls or native-tls feature to be enabled. (I’m on macOS Tahoe, running on an ARM64 machine.) Rather than struggling with building a static lib for openblas, I wanted to use native macOS BLAS and LAPACK libraries. To use macOS Accelerate framework instead of openblas, we must change the spec file (Cargo.toml) so that it looks like:

[dependencies]
csv = "1.1"
ndarray = {version = "0.15", features=["rayon", "serde", "blas"]}
ndarray-linalg = {version = "0.16", default-features = false}
blas-src = { version = "0.8", features = ["accelerate"] }

We also need to add specific instructions for linking the target in build.rs.

fn main() {
 println!("cargo:rustc-link-lib=framework=Accelerate")
}

This is discussed in the following GH issue.

Running the original code as is, everything works out of the box:

 -%<--
 Compiling ndarray-linalg v0.16.0
 Finished `dev` profile [unoptimized + debuginfo] target(s) in 14.93s
 Running `target/debug/regress`
[2.9388893694493694, 0.04576464545542497, 0.188530016918319, -0.0010374930424147009]
RSE: 1.6769760888385676

Full code available on GH in this subfolder, with all credits to Lorenzo Celli for the original code.

♪ A Flock of Seagulls • Over My Head

Rust on aliquote.org

Linear regression in Rust