A toy probabilistic programming library written in Rust for learning about Bayesian inference. This is a hobby project to explore basic concepts in probabilistic modeling and MCMC sampling.
- Basic Distributions: Normal, Exponential, and Inverse-Gamma with sampling and log-probability evaluation.
- Simple Model Definition: Define models using closures that return log-probabilities.
- Metropolis-Hastings Sampler: Basic MCMC implementation for posterior sampling.
- Sample Analysis: Compute means, standard deviations, and quantiles from MCMC output.
- Examples: A few working examples of Bayesian inference problems.
use ppf::*;
// Define data
let data = [2.1, 1.9, 2.0, 2.2, 1.8];
// Create Bayesian model
let model = Model::new(move |params: &[f64]| {
let mu = params[0];
let prior = Normal::new(0.0, 10.0).log_prob(mu);
let likelihood: f64 = data.iter()
.map(|&x| Normal::new(mu, 1.0).log_prob(x))
.sum();
prior + likelihood
});
// Run MCMC inference
let mh = MetropolisHastings::new(1, 0.5);
let samples = mh.sample(&model, 10_000);
// Analyze results
println!("Posterior mean: {:.3}", samples.mean(0));
println!("95% CI: [{:.3}, {:.3}]",
samples.quantile(0, 0.025), samples.quantile(0, 0.975));The library has four simple modules:
Three basic probability distributions:
- Normal: Gaussian distribution
- Exponential: Simple exponential distribution
- InverseGamma: Inverse-gamma distribution (useful for variance priors)
Simple model wrapper that takes a closure:
let model = Model::new(|params: &[f64]| {
// Return log probability given parameters
prior_log_prob + likelihood_log_prob
});Basic Metropolis-Hastings implementation for MCMC.
Container for MCMC samples with basic statistics:
let posterior_mean = samples.mean(0);
let credible_interval = (samples.quantile(0, 0.025), samples.quantile(0, 0.975));use ppf::distributions::*;
use rand::rng;
// Sample from distributions
let normal = Normal::new(0.0, 1.0);
let exponential = Exponential::new(1.5);
let inv_gamma = InverseGamma::new(2.0, 1.0);
let mut rng = rng();
println!("Normal sample: {:.3}", normal.sample(&mut rng));
println!("Exponential sample: {:.3}", exponential.sample(&mut rng));
println!("InverseGamma sample: {:.3}", inv_gamma.sample(&mut rng));
// Evaluate log probabilities
println!("Normal log-prob at 0: {:.3}", normal.log_prob(0.0));
println!("Exponential log-prob at 1: {:.3}", exponential.log_prob(1.0));Run the examples to see the library in action:
# Simple Bayesian inference (unknown mean)
cargo run --example bayesian_inference
# Joint inference of mean and variance (exponential prior)
cargo run --example infer_mean_and_variance
# Joint inference with inverse-gamma prior on variance
cargo run --example infer_mean_variance_invgamma
# Basic distribution sampling
cargo run --example demo# Build the project
cargo build
# Run tests
cargo test
# Check code style
cargo clippyThis is a learning project with many limitations:
- Only three basic distributions.
- Just one MCMC algorithm (Metropolis-Hastings).
- No convergence diagnostics beyond acceptance rate.
- Limited to continuous parameters.
- No automatic differentiation or gradient-based methods.
- Not optimized for performance.
For real work, consider mature libraries like PyMC, Stan, or Turing.jl.
MIT License - see LICENSE file for details.