Struct EvolutionStrategy 

pub struct EvolutionStrategy<B: Backend> { /* private fields */ }

Classical Evolution Strategy.

Example

use burn::backend::Flex;
use rlevo_evolution::algorithms::es_classical::{EsConfig, EsKind, EvolutionStrategy};

let strategy = EvolutionStrategy::<Flex>::new();
let params = EsConfig::default_for(EsKind::MuPlusLambda { mu: 5, lambda: 20 }, 10);
let _ = (strategy, params);

Implementations

impl<B: Backend> EvolutionStrategy<B>

pub fn new() -> Self

Builds a new (stateless) strategy object.

Trait Implementations

impl<B: Clone + Backend> Clone for EvolutionStrategy<B>

fn clone(&self) -> EvolutionStrategy<B>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<B: Copy + Backend> Copy for EvolutionStrategy<B>

impl<B: Debug + Backend> Debug for EvolutionStrategy<B>

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

Formats the value using the given formatter.

impl<B: Default + Backend> Default for EvolutionStrategy<B>

fn default() -> EvolutionStrategy<B>

Returns the “default value” for a type.

impl<B: Backend> Strategy<B> for EvolutionStrategy<B>

where B::Device: Clone,

fn init( &self, params: &EsConfig, rng: &mut dyn Rng, device: &<B as BackendTypes>::Device, ) -> EsState<B>

Samples the initial parent population uniformly from params.bounds via a deterministic seed_stream (host-RNG convention) and initializes all σ values to params.initial_sigma.

fn ask( &self, params: &EsConfig, state: &EsState<B>, rng: &mut dyn Rng, device: &<B as BackendTypes>::Device, ) -> (Tensor<B, 2>, EsState<B>)

Generates the offspring population for the current generation.

On the very first call (before any tell), returns the initial parents unchanged so that they can be fitness-evaluated as the seed population. On subsequent calls, duplicates parents by uniform random selection, applies log-normal σ adaptation (multi-parent variants) or inherits the shared σ ((1+1) / (1+λ)), then mutates via per-individual Gaussian noise. All stochastic draws go through seed_stream (host-RNG convention); offspring σ values are appended to state.sigmas for consumption by tell.

fn tell( &self, params: &EsConfig, offspring: Tensor<B, 2>, fitness: Tensor<B, 1>, state: EsState<B>, _rng: &mut dyn Rng, ) -> (EsState<B>, StrategyMetrics)

Applies variant-specific selection and σ adaptation, then returns the updated state and a per-generation metrics snapshot.

Variant behaviour:

• (1+1): greedy replacement; σ updated by Rechenberg’s 1/5th success rule every 10·D steps.
• (1+λ): best offspring replaces the parent only if it strictly improves fitness; σ is carried over unchanged.
• (μ,λ): selects the μ best offspring; parent pool discarded. Survivor σ values are gathered by the same truncation indices.
• (μ+λ): selects the μ best of the combined parent + offspring pool. Survivor σ values are drawn from the concatenated σ vector by the same indices.

The first tell after init bootstraps parent_fitness from the initial-population evaluation rather than running selection.

fn best(&self, state: &EsState<B>) -> Option<(Tensor<B, 2>, f32)>

Returns the best-so-far genome and its fitness, or None before the first tell call.

type Params = EsConfig

Static parameters for a run.

type State = EsState<B>

Generation-to-generation state.

type Genome = Tensor<B, 2>

Genome container produced by ask.