Skip to main content

SearchAlgorithm

wafrift_evolution::search

Trait SearchAlgorithm 

Source 
pub trait SearchAlgorithm:
    Send
    + Sync
    + Debug {
    // Required methods
    fn name(&self) -> &'static str;
    fn initialize(
        &mut self,
        population: Vec<Chromosome>,
        gene_pool: &GenePool,
        rng: &mut StdRng,
    );
    fn request_evaluations(
        &mut self,
        n: usize,
        rng: &mut StdRng,
    ) -> Vec<EvalCandidate>;
    fn submit_evaluations(&mut self, results: Vec<(u64, OracleVerdict)>);
    fn should_terminate(&self, stats: &SearchStats, budget: &Budget) -> bool;
    fn best(&self) -> Option<&Chromosome>;
    fn checkpoint(&self) -> Result<Vec<u8>, EvolutionError>;
    fn restore(&mut self, bytes: &[u8]) -> Result<(), EvolutionError>;

    // Provided methods
    fn population_snapshot(&self) -> Vec<Chromosome> { ... }
    fn clone_box(&self) -> Box<dyn SearchAlgorithm> { ... }
}
Expand description

Core trait implemented by all search algorithms.

Each algorithm manages its own internal state (population, archive, temperature, tabu list, etc.). The EvolutionEngine handles caching, budgeting, and batching on top of this trait.

Required Methods§

Source

fn name(&self) -> &'static str

Algorithm name.

Source

fn initialize( &mut self, population: Vec<Chromosome>, gene_pool: &GenePool, rng: &mut StdRng, )

Initialize the algorithm with a seed population.

Source

fn request_evaluations( &mut self, n: usize, rng: &mut StdRng, ) -> Vec<EvalCandidate>

Request up to n candidates for parallel evaluation.

Returns candidates with stable IDs. The caller evaluates them and later calls submit_evaluations.

Source

fn submit_evaluations(&mut self, results: Vec<(u64, OracleVerdict)>)

Submit evaluation results.

The ID in each tuple must match an ID previously returned by request_evaluations.

Source

fn should_terminate(&self, stats: &SearchStats, budget: &Budget) -> bool

Check whether the algorithm thinks search should stop.

Source

fn best(&self) -> Option<&Chromosome>

Get the best chromosome found so far.

Source

fn checkpoint(&self) -> Result<Vec<u8>, EvolutionError>

Serialize internal state to bytes.

Source

fn restore(&mut self, bytes: &[u8]) -> Result<(), EvolutionError>

Restore internal state from bytes.

Provided Methods§

Source

fn population_snapshot(&self) -> Vec<Chromosome>

Snapshot the algorithm’s current “live” chromosomes — the set the engine is actively searching from.

Population-based algorithms (NoveltySearch, MapElites) return their full pool. Single-state algorithms (HillClimbing, SimulatedAnnealing, TabuSearch) return the singleton current/best so the engine sees len() == 1 and reports “no pairwise diversity yet” rather than zero.

Used by EvolutionEngine::diversity_score to drive adaptive mutation pressure. Cloning is acceptable here because callers run it at engine-tick rate, not per-evaluation.

Source

fn clone_box(&self) -> Box<dyn SearchAlgorithm>

Deep-clone this algorithm into a fresh trait object.

The default implementation falls back to a checkpoint → restore round-trip via serde_json — correct for any algorithm that implements those, but slow on large grids / archives because every chromosome is JSON-serialised.

Concrete algorithms override this with a direct in-memory Clone to bypass the serde round-trip — typically 10-100× faster on populated state. The override is what EvolutionEngine::clone uses on the proxy path, where allocation spikes from JSON were the original blocker.

Implementors§