Module inference 

Evolution as inference: the PPL-native layer (requires the ppl feature)

This module makes “evolutionary algorithms as probabilistic programs” literal. Given a fitness f and a prior program p(x) over genomes (a prior::GenomePrior — any fugue Model<G>), the Boltzmann/Gibbs posterior

    π_β(x) ∝ p(x) · exp(β · f(x))

is itself a fugue program (prior.model().bind(|g| factor(β·f(g)))), and every sampler here is fugue’s own inference machinery run against it:

• model::EvolutionModel assembles the target program; all densities are obtained by running/replaying it (no hand-written density code).
• mh::EvolutionChain delegates to fugue::adaptive_single_site_mh — typed proposals move every site kind (Bool/U64/Usize/I64/F64), fixing the historical F64-only dead-chain bug.
• smc::EvolutionSMC delegates to fugue::adaptive_smc_with_kernel: adaptive likelihood-tempering (β applied exactly once), ESS-driven resampling, per-particle MH rejuvenation, an optional population-coupled crossover kernel, and an unbiased log-evidence estimate.
• bayesian_ga::BayesianAdaptiveGA keeps the Thompson-sampling operator selection (conjugate Beta/Gamma posteriors, now backed by rand_distr).

effect_handlers retains the genuine fugue::Handler implementations (TraceScoringHandler, RecordingHandler) and the operator observation hooks; trace_operators retains the value-level trace operators.

See examples/bayesian_evolution.rs for an end-to-end pipeline.

Modules

bayesian_ga

A single-level Bayesian adaptive genetic algorithm

effect_handlers

Fugue Handlers and operation hooks for evolutionary operators

grammar

Expression trees as probabilistic grammars: genetic programming as exact Bayesian inference

likelihood

Likelihoods as programs

mh

Metropolis–Hastings over the Boltzmann target, delegated to fugue

model

The Boltzmann posterior over genomes, as a fugue program

pareto

Multi-objective optimization as Bayesian inference: the Pareto posterior

prelude

prior

Priors over genomes as probabilistic programs

smc

Tempered SMC over the Boltzmann posterior, built on fugue’s SMC engine

trace_operators

Trace-based genetic operators