Struct CmaEsConfig 

pub struct CmaEsConfig {

    pub pop_size: usize,
    pub genome_dim: usize,
    pub bounds: Bounds,
    pub initial_sigma: f32,
    pub mu: usize,
    pub weights: Vec<f32>,
    pub mu_eff: f32,
    pub c_sigma: f32,
    pub d_sigma: f32,
    pub c_c: f32,
    pub c_1: f32,
    pub c_mu: f32,
    pub chi_n: f32,
}

Static configuration for a CMA-ES run.

Construct with CmaEsConfig::default_for (derives λ from the dimension per Hansen 2016) or CmaEsConfig::with_pop_size (explicit λ, e.g. a larger population for multimodal landscapes). The recombination weights and learning rates are all derived from (λ, D) and cached as fields so Strategy::tell reads them without recomputing.

Fields

pop_size: usize

Offspring population size λ.

genome_dim: usize

Genome dimensionality D.

bounds: Bounds

Search-space bounds; used only to sample the initial mean m⁰. Offspring are not clamped (CMA-ES samples in unbounded ℝᴰ).

initial_sigma: f32

Initial global step size σ.

mu: usize

Number of selected parents μ = ⌊λ/2⌋.

weights: Vec<f32>

Recombination weights wᵢ (length μ, positive, summing to 1).

mu_eff: f32

Variance-effective selection mass μ_eff = 1 / Σ wᵢ².

c_sigma: f32

CSA learning rate c_σ.

d_sigma: f32

CSA damping d_σ.

c_c: f32

Anisotropic-path learning rate c_c.

c_1: f32

Rank-1 covariance learning rate c_1.

c_mu: f32

Rank-μ covariance learning rate c_μ.

chi_n: f32

Expected length of N(0, I), χ_n ≈ √D (1 − 1/4D + 1/21D²).

Implementations

impl CmaEsConfig

pub fn default_for(genome_dim: usize) -> Self

Default configuration for dimensionality D, with the Hansen-2016 population λ = 4 + ⌊3 ln D⌋.

Sets bounds = (-5.12, 5.12) (the standard Sphere/Rastrigin domain) and initial_sigma = 1.0.

pub fn with_pop_size(pop_size: usize, genome_dim: usize) -> Self

Configuration with an explicit population size λ.

Larger λ improves basin-finding on multimodal landscapes (Hansen 2016, §A); all derived weights and learning rates follow from (λ, D).

The pop_size ≥ 2 invariant is enforced by Validate::validate at the harness chokepoint, not by this infallible producer.

Trait Implementations

impl Clone for CmaEsConfig

fn clone(&self) -> CmaEsConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CmaEsConfig

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

Formats the value using the given formatter.

impl Validate for CmaEsConfig

fn validate(&self) -> Result<(), ConfigError>

Fail-fast: reports the first violation, derived from validate_all so the two never disagree.

fn validate_all(&self) -> Result<(), Vec<ConfigError>>

Accumulate-all: reports every violated invariant in one pass.

Unlike most configs, CmaEsConfig exposes its derived fields — recombination weights, mu_eff, and the five learning rates — as public struct fields (so callers can construct one by hand). The default_for / with_pop_size producers keep them mutually consistent, but a hand-built literal can desync several at once; listing all violations then beats fixing them one recompile at a time.