ordofp_bayes - Probabilistic Programming for OrdoFP

"Probabilitas est ratio incertitudinis." — Probability is the measure of uncertainty. (Modern)

This crate provides Bayesian inference algorithms for probabilistic programming, using functional abstractions, as part of the OrdoFP ecosystem.

Adapted third-party algorithms are inventoried in ORIGINAL_SOURCE.md in this crate and the repo-root THIRD_PARTY_NOTICES.md.

Features

• Sequential Monte Carlo (SMC)
• Metropolis-Hastings (MCMC)
• Importance Sampling
• Parallel particle generation via Rayon (rayon feature)

Performance

Weight calculations are written as tight slice loops that LLVM auto-vectorizes; there is no explicit SIMD (std::simd) code in this crate.

Example

use ordofp_bayes::distributions::Normal;
use ordofp_bayes::{Distribution, MetropolisHastings};
use rand::SeedableRng;
use rand::rngs::StdRng;
use std::sync::atomic::{AtomicU64, Ordering};

// Give every closure call its own seed, so the 1000 draws below are
// genuinely independent samples rather than 1000 copies of one number.
static SEED: AtomicU64 = AtomicU64::new(1);
fn next_seed() -> u64 {
    SEED.fetch_add(1, Ordering::Relaxed)
}

// Define a simple model: sample once from a prior distribution.
let mut rng = StdRng::seed_from_u64(42);
// Note: `MetropolisHastings::infer` draws `iterations` independent
// samples via resampling rather than running an MCMC chain, so
// `burn_in` (the `100` below) has no effect on this code path — it
// only matters for `infer_traceable`'s trace-based chain.
let mh = MetropolisHastings::new(1000, 100);

// Run inference
let samples: Vec<f64> = mh.infer(
    || {
        let mut local_rng = StdRng::seed_from_u64(next_seed());
        Normal::new(0.0, 1.0).sample(&mut local_rng)
    },
    &mut rng,
);
assert_eq!(samples.len(), 1000);
// Real sampling variation: 1000 draws from a continuous Normal are not
// all identical (this cannot flake — the odds of a false failure are
// astronomically small).
assert!(samples.windows(2).any(|w| w[0] != w[1]));