Struct HyperparameterOptimizer 

pub struct HyperparameterOptimizer<T: Float> { /* private fields */ }

Hyperparameter optimizer for NAS and discovered architectures

Implementations

impl<T: Float + Default + Clone + Send + Sync> HyperparameterOptimizer<T>

pub fn new( search_space: HyperparameterSpace<T>, strategy: OptimizationStrategy, ) -> Self

Create a new hyperparameter optimizer, seeded from OS entropy.

pub fn with_seed( search_space: HyperparameterSpace<T>, strategy: OptimizationStrategy, seed: u64, ) -> Self

Create a fully reproducible optimizer.

pub fn suggest_configuration( &mut self, ) -> Result<HyperparameterConfiguration<T>>

Generate the next hyperparameter configuration to evaluate.

Returns an error for a strategy this module does not implement (OptimizationStrategy::ParticleSwarm, SuccessiveHalving, Hyperband, BOHB) rather than silently performing random search, and for a search space that declares no parameters at all.

The model-based strategies (Bayesian, TPE, Evolutionary) need a random initial design first. During that phase they sample from the prior and set HyperparameterOptimizer::last_strategy_used to OptimizationStrategy::Random, so the caller can see which algorithm actually produced the suggestion.

pub fn last_strategy_used(&self) -> OptimizationStrategy

Which strategy produced the most recent suggestion. Differs from HyperparameterOptimizer::strategy exactly while a model-based strategy is still running its random initial design.

pub fn strategy(&self) -> OptimizationStrategy

The configured strategy.

pub fn grid_resolution(&self) -> usize

Number of points per continuous grid axis.

pub fn set_grid_resolution(&mut self, resolution: usize)

Set the number of points per continuous grid axis.

pub fn grid_size(&self) -> Option<usize>

Total number of distinct grid points, or None if the product overflows.

pub fn set_acquisition_function(&mut self, acquisition: AcquisitionFunction)

Choose the acquisition function used by the Bayesian strategy.

pub fn set_tpe_config(&mut self, config: TpeConfig)

Override the TPE tunables.

pub fn set_surrogate_config(&mut self, config: SurrogateConfig)

Override the Bayesian surrogate tunables.

pub fn set_evolution_config(&mut self, config: EvolutionConfig)

Override the evolutionary tunables.

pub fn get_search_space(&self) -> &HyperparameterSpace<T>

The search space this optimizer explores.

pub fn population(&self) -> &[HyperparameterConfiguration<T>]

The evolutionary population (fittest first), which is now genuinely read by OptimizationStrategy::Evolutionary rather than being write-only state.

pub fn surrogate_model(&self) -> Option<&SurrogateModel<T>>

The fitted surrogate’s recorded training data, or None before the Bayesian strategy has run.

pub fn record_evaluation(&mut self, evaluation: HyperparameterEvaluation<T>)

Record evaluation result

pub fn should_stop_early(&self) -> bool

Check if optimization should stop early

pub fn get_best_configuration(&self) -> Option<&HyperparameterConfiguration<T>>

Get the best configuration found so far

pub fn get_statistics(&self) -> OptimizationStatistics<T>

Get optimization statistics

Trait Implementations

impl<T: Debug + Float> Debug for HyperparameterOptimizer<T>

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

Formats the value using the given formatter.