Struct EvolutionPosterior 

pub struct EvolutionPosterior<G: TraceGenome> {
    pub particles: Vec<Particle>,
    pub log_evidence: f64,
    /* private fields */
}

The result of a tempered-SMC evolution run: fugue particles (traces + normalized weights) approximating the Boltzmann posterior π ∝ p·exp(f), plus the log-evidence estimate.

Genomes are not cached on particles; they are recovered by decode-replay (replaying the particle’s trace through the prior/target program, whose return value is the decoded genome).

Fields

particles: Vec<Particle>

Final weighted particle population (fugue particles).

log_evidence: f64

Unbiased estimate of the log normalizing constant log Σ_x p(x)·exp(f(x)) — the Bayesian model score.

Implementations

impl<G: TraceGenome> EvolutionPosterior<G>

pub fn genome(&self, particle: &Particle, model_fn: &impl Fn() -> Model<G>) -> G

Recover the genome of one particle by replaying its trace.

pub fn genomes(&self, model_fn: &impl Fn() -> Model<G>) -> Vec<(G, f64)>

Decode the whole population as (genome, normalized_weight) pairs.

pub fn weighted_mean(&self, coord: usize) -> f64

Self-normalised weighted posterior mean of coordinate gene#coord.

pub fn weighted_variance(&self, coord: usize) -> f64

Self-normalised weighted posterior variance of coordinate gene#coord.

pub fn best<F>( &self, fitness: &F, model_fn: &impl Fn() -> Model<G>, ) -> Option<(G, f64)>

where F: Fitness<Genome = G, Value = f64>,

The decoded genome with the highest fitness, and that fitness — the optimizer-mode readout for benchmarking against the classic layer.