Crate ndarray_glm[][src]

Expand description

A rust library for performing GLM regression with data represented in ndarrays. The ndarray-linalg crate is used to allow optimization of linear algebra operations with BLAS.

This crate is early alpha and may change rapidly. No guarantees can be made about the accuracy of the fits.

Feature summary:

  • Linear, logistic, Poisson, and binomial regression (more to come)
  • Generic over floating-point type
  • L2 (ridge) regularization
  • Statistical tests of fit result
  • Alternative and custom link functions

Setting up BLAS backend

See the backend features of ndarray-linalg for a description of the available BLAS configuartions. You do not need to include ndarray-linalg in your crate; simply provide the feature you need to ndarray-glm and it will be forwarded to ndarray-linalg.

Examples using OpenBLAS are shown here. In principle you should also be able to use Netlib or Intel MKL, although these backends are untested.

Ensure that the development OpenBLAS library is installed on your system. In Debian/Ubuntu, for instance, this means installing libopenblas-dev. Then, put the following into your crate’s Cargo.toml:

ndarray = { version = "0.15", features = ["blas"]}
ndarray-glm = { version = "0.0.10", features = ["openblas-system"] }

Compile OpenBLAS from source

This option does not require OpenBLAS to be installed on your system, but the initial compile time will be very long. Use the folling lines in your crate’s Cargo.toml.

ndarray = { version = "0.15", features = ["blas"]}
ndarray-glm = { version = "0.0.10", features = ["openblas-static"] }


Basic linear regression:

use ndarray_glm::{array, Linear, ModelBuilder};

let data_y = array![0.3, 1.3, 0.7];
let data_x = array![[0.1, 0.2], [-0.4, 0.1], [0.2, 0.4]];
let model = ModelBuilder::<Linear>::data(&data_y, &data_x).build().unwrap();
let fit =;
// The result is a flat array with the first term as the intercept.
println!("Fit result: {}", fit.result);

Data standardization and L2 regularization:

use ndarray_glm::{array, Linear, ModelBuilder, utility::standardize};

let data_y = array![0.3, 1.3, 0.7];
let data_x = array![[0.1, 0.2], [-0.4, 0.1], [0.2, 0.4]];
// The design matrix can optionally be standardized, where the mean of each independent
// variable is subtracted and each is then divided by the standard deviation of that variable.
let data_x = standardize(data_x);
let model = ModelBuilder::<Linear>::data(&data_y, &data_x).build().unwrap();
// L2 (ridge) regularization can be applied with l2_reg().
let fit = model.fit_options().l2_reg(1e-5).fit().unwrap();
println!("Fit result: {}", fit.result);

Logistic regression with a non-canonical link function:

use ndarray_glm::{array, Logistic, logistic_link::Cloglog, ModelBuilder};

let data_y = array![true, false, false, true, true];
let data_x = array![[0.5, 0.2], [0.1, 0.3], [0.2, 0.6], [0.6, 0.3], [0.4, 0.4]];
let model = ModelBuilder::<Logistic<Cloglog>>::data(&data_y, &data_x).build().unwrap();
let fit = model.fit_options().l2_reg(1e-5).fit().unwrap();
println!("Fit result: {}", fit.result);


pub use model::ModelBuilder;


define the error enum for the result of regressions

Defines traits for link functions

Link functions for logistic regression

Collect data for and configure a model

numerical trait constraints

utility functions for internal library use


Create an Array with one, two or three dimensions.


Binomial regression with a fixed N. Non-canonical link functions are not possible at this time due to the awkward ergonomics with the const trait parameter N.

the result of a successful GLM fit

Linear regression with constant variance (Ordinary least squares).

Logistic regression

Poisson regression over an unsigned integer type.

Type Definitions

one-dimensional array

two-dimensional array

one-dimensional array view

two-dimensional array view