Struct WhaleOptimization 

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

Whale Optimization Algorithm strategy.

Panics

Strategy::ask panics if called a second time without an intervening Strategy::tell — specifically, if state.best_genome is None after the first generation. In normal harness-driven usage this cannot happen: ask on the first call returns the initial positions unevaluated; tell then sets best_genome; and every subsequent ask finds it populated. Bypassing the harness and calling ask twice in a row without a tell in between will trigger the assert.

Example

use burn::backend::Flex;
use rlevo_evolution::algorithms::metaheuristic::woa::{WhaleOptimization, WoaConfig};

let strategy = WhaleOptimization::<Flex>::new();
let params = WoaConfig::default_for(32, 10);
let _ = (strategy, params);

Implementations

impl<B: Backend> WhaleOptimization<B>

pub fn new() -> Self

Builds a new (stateless) strategy object.

Trait Implementations

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

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

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

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

Formats the value using the given formatter.

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

fn default() -> WhaleOptimization<B>

Returns the “default value” for a type.

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

where B::Device: Clone,

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

Samples the initial whale positions uniformly within WoaConfig::bounds using the host-RNG convention and sets the generation counter to zero.

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

Proposes the next whale positions.

On the first call returns the initial positions unchanged. On subsequent calls, each whale independently selects one of three moves based on host-sampled scalars p and |A|:

• p < 0.5, |A| < 1 — encircle the current best,
• p < 0.5, |A| ≥ 1 — search toward a random other whale,
• p ≥ 0.5 — spiral toward the current best.

The three candidate tensors are computed in parallel and composed with boolean masks; no divergent kernel paths are used. Results are clamped to WoaConfig::bounds.

fn tell( &self, _params: &WoaConfig, population: Tensor<B, 2>, fitness: Tensor<B, 1>, state: WoaState<B>, _rng: &mut dyn Rng, ) -> (WoaState<B>, StrategyMetrics)

Records evaluated fitness, updates the best-so-far (food source), and increments the generation counter.

Returns the updated WoaState and a StrategyMetrics snapshot for the completed generation.

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

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

type Params = WoaConfig

Static parameters for a run.

type State = WoaState<B>

Generation-to-generation state.

type Genome = Tensor<B, 2>

Genome container produced by ask.