#markov-chain #monte-carlo #sampling #mcmc #random

metromc

Markov chain Monte Carlo sampling using the Independence Metropolis-Hastings algorithm

2 unstable releases

0.2.0 Jan 4, 2024
0.1.0 Jan 4, 2024

#1822 in Algorithms

MIT license

11KB
189 lines

metromc

Markov chain Monte Carlo (MCMC) sampling using the Independence Metropolis-Hastings algorithm with uniform transition kernel.

Crates.io docs.rs Build Status

Uses the tinyrand RNG to sample at a rate of ~50M samples/sec.

Supports the following distributions:

It is easy to add more univariate distributions by supplying an implementation of a PDF or wrapping one from the excellent statrs crate.

Example

Draw samples from the Gaussian distribution using MCMC.

use std::ops::RangeInclusive;
use tinyrand::Wyrand;
use metromc::gaussian::Gaussian;
use metromc::sampler::{Config, Sampler};

// sample from Gaussian with µ=0.0 and σ=1.0, in the interval [-5.0, 5.0]
let sampler = Sampler::new(Config {
    rand: Wyrand::default(),
    dist: Gaussian::new(0.0, 1.0),
    range: -5.0..=5.0,
});

// take 1,000 samples after dropping the first 10
for sample in sampler.skip(10).take(1_000) {
    println!("{sample:.6}");
}

Dependencies

~5MB
~97K SLoC