A pure Typst library for common probability distributions. Each distribution module provides a constructor, PMF/PDF and CDF. Some modules also include inverse-transform sampling and mean/variance fields.

This library is still in early development, please use with caution. Typst is not designed for robust statistical use, and built-in functions like calc.binom overflow quickly. For any missing pieces, or feature requests, please open an Issue!

Basic usage

#import "@preview/distro:0.2.0": normal, binomial, bernoulli
#import calc: sqrt, pi
#import "@preview:distro": normal

#{
  let Z = normal.new(mean: 0, std: 1)
  assert((Z.mean, Z.variance) == (0, 1))
  assert(normal.pdf(Z)(0) == 1 / sqrt(2 * pi))
}

Sampling random variates

Provide a uniform random variate from a PRNG such as suiji and distro will sample a random variate from your distribution of choice:

#import "@preview/distro:0.2.0": categorical
#import "@preview/suiji:0.5.1": gen-rng-f, uniform-f

#let Cat = categorical.new((0.1, 0.3, 0.2, 0.4))

// Random variate generation
#let n_samples = 1000
#let counts = (0,) * Cat.weights.len()
#let (rng, u) = (gen-rng-f(42), none)
#for _ in range(n_samples) {
  (rng, u) = uniform-f(rng)
  let result = categorical.sample(Cat, u)
  counts.at(result) += 1
}

// Frequency Table
#table(
  columns: (auto, auto, auto),
  inset: 10pt,
  align: center,
  [*Category*], [*Count*], [*%*],
  ..for (i, count) in counts.enumerate() {
    (
      [#i],
      [#count],
      [#(calc.round(count / n_samples * 100, digits: 1))%],
    )
  },
)

See examples/sampling.typ for further examples.

Distributions

Click an image to see the source.

Discrete

Bernoulli	Binomial
Categorical	Geometric
Poisson	Discrete Uniform

Continuous

Beta	Continuous Uniform
Exponential	Gamma
Normal

Changelog

v0.2.0

• Minimum Typst version increased to v0.15.0 to accommodate replacement of package-defined functions/erf.typ error-function approximation with Typst-vendored new builtin calc.erf (via libm).
• Re-compiled the distribution plot gallery SVGs to leverage the minified SVG export (reduced gallery size by 200K).

v0.1.0

Initial package creation!

Acknowledgements

Numerical algorithms for the gamma and beta functions are adapted from the Rust statrs crate (MIT licensed).