Struct StrategyMetrics 

pub struct StrategyMetrics { /* private fields */ }

Per-generation summary reported by Strategy::tell.

All statistics refer to the generation that just finished evaluating. These values are in canonical (maximise) space: higher is better. Strategies are sense-unaware, so the metrics they emit are always canonical. EvolutionaryHarness::latest_metrics maps them back to the objective’s declared sense before surfacing them to callers, so a Minimize landscape reads as its natural cost (Sphere → 0).

When printed in a benchmark showcase (e.g. ackley_showcase), the two most informative fields are:

• best_fitness_ever — the best (canonical: largest) fitness seen across all generations so far. This is a rolling maximum that tells you how close the best individual ever found came to the optimum.
• mean_fitness — the arithmetic mean of the current generation’s per-individual fitness vector. This tells you the average quality of the population in that generation.

A large gap between best_fitness_ever and mean_fitness in the final generation usually indicates premature convergence: a few elite individuals found a good basin while the rest of the population is still scattered. A small gap suggests the whole population has settled near the same optimum.

Implementations

impl StrategyMetrics

pub fn from_host_fitness( generation: usize, fitnesses: &[f32], best_fitness_ever: f32, ) -> Self

Computes population statistics from a host-side fitness slice.

Each value is passed through the crate’s fitness-hygiene primitive sanitize_fitness before folding, so NaN → −∞ and +∞ → f32::MAX (the maximise convention, ADR 0023/0034) consistently across every statistic — best/worst can no longer silently drop a NaN (comparisons against NaN are false) while the sum propagates it.

mean_fitness is computed over the finite members only: a sanitized −∞ member (a NaN evaluation, or a genuine worst-sentinel) is excluded from the average and counted in broken_count instead (ADR 0034). This keeps a single broken individual from blanking the whole mean to −∞ while still surfacing that the population is unhealthy. +∞ → f32::MAX members are finite and are included, so an optimal individual cannot blow the mean to +∞. If every member is broken, mean_fitness = −∞ (degenerate but well-defined).

Panics

Panics if fitnesses is empty. Callers hold a non-empty population by construction — pop_size is validated non-zero at the harness constructor (ADR 0026).

pub fn generation(&self) -> usize

Zero-based generation index.

pub fn population_size(&self) -> usize

Number of individuals evaluated in this generation.

pub fn best_fitness(&self) -> f32

Best (canonical: largest) fitness observed in this generation.

pub fn mean_fitness(&self) -> f32

Mean fitness across this generation’s population.

Averaged over the finite members only; broken (−∞) members are excluded and reported by broken_count (ADR 0034).

pub fn broken_count(&self) -> usize

Number of non-finite (broken) individuals excluded from mean_fitness this generation (ADR 0034).

Zero on a healthy run; non-zero flags a population carrying NaN/−∞ members.

pub fn worst_fitness(&self) -> f32

Worst (canonical: smallest) fitness observed in this generation.

pub fn best_fitness_ever(&self) -> f32

Best (canonical: largest) fitness seen across all generations to date.

Trait Implementations

impl Clone for StrategyMetrics

fn clone(&self) -> StrategyMetrics

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for StrategyMetrics

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.