Struct FireflyAlgorithm 

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

Firefly Algorithm strategy.

When the custom-kernels feature is off, FireflyConfig enforces a pop_size <= FIREFLY_PURE_TENSOR_CAP (= 128) cap through Validate::validate at the harness chokepoint, since the pure-tensor path materializes an (N, N, D) pairwise tensor. With the feature on the same cap is surfaced via a debug_assert! in Strategy::init, because the fused kernel super::kernels::pairwise_attract_cube is still designed-only and the strategy keeps using the pure-tensor path in the meantime.

Example

use burn::backend::Flex;
use rlevo_evolution::algorithms::metaheuristic::firefly::{FireflyAlgorithm, FireflyConfig};

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

Implementations

impl<B: Backend> FireflyAlgorithm<B>

pub fn new() -> Self

Builds a new (stateless) strategy object.

Trait Implementations

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

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

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

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

Formats the value using the given formatter.

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

fn default() -> FireflyAlgorithm<B>

Returns the “default value” for a type.

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

where B::Device: Clone,

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

Build the initial swarm by host-sampling pop_size positions uniformly in [bounds.lo, bounds.hi].

Positions are drawn from a deterministic seed_stream so initialisation is bit-stable regardless of core count or test ordering; the process-wide Flex RNG is never touched.

The pop_size <= FIREFLY_PURE_TENSOR_CAP cap (without custom-kernels) is enforced by FireflyConfig’s Validate impl at the harness chokepoint.

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

Propose the next swarm positions.

On the first call (state.fitness is empty) returns the initial positions unchanged so the caller can evaluate generation zero. On subsequent calls, computes the pairwise attractiveness update via pure_tensor_attract and clips positions to params.bounds. The noise seed is derived from the host RNG through seed_stream, keeping draws reproducible.

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

Ingest fitness values, update the swarm, and advance the generation counter.

Pulls fitness to host, updates state.positions and state.fitness, then refreshes the best-so-far genome if the current generation contains a new maximum. Returns the updated state and a StrategyMetrics snapshot for the completed generation.

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

Returns the best-so-far (genome, fitness) pair, or None before the first tell call.

type Params = FireflyConfig

Static parameters for a run.

type State = FireflyState<B>

Generation-to-generation state.

type Genome = Tensor<B, 2>

Genome container produced by ask.