Struct BayesianOptimizer 

pub struct BayesianOptimizer { /* private fields */ }

The Bayesian optimizer.

Supports sequential single-objective, batch, multi-objective (ParEGO), and constrained optimization.

Implementations

impl BayesianOptimizer

pub fn new( bounds: Vec<(f64, f64)>, config: BayesianOptimizerConfig, ) -> OptimizeResult<Self>

Create a new Bayesian optimizer.

Arguments

• bounds - Search bounds for each dimension: [(lo, hi), ...]
• config - Optimizer configuration

pub fn with_kernel( bounds: Vec<(f64, f64)>, kernel: Box<dyn SurrogateKernel>, config: BayesianOptimizerConfig, ) -> OptimizeResult<Self>

Create a new optimizer with a custom kernel.

pub fn add_constraint<F>(&mut self, name: &str, func: F)

where F: Fn(&ArrayView1<'_, f64>) -> f64 + Send + Sync + 'static,

Add a constraint: satisfied when g(x) <= 0.

pub fn warm_start( &mut self, x_data: &Array2<f64>, y_data: &Array1<f64>, ) -> OptimizeResult<()>

Warm-start from previous evaluations.

pub fn optimize<F>( &mut self, objective: F, n_iter: usize, ) -> OptimizeResult<BayesianOptResult>

where F: Fn(&ArrayView1<'_, f64>) -> f64,

Run the sequential optimization loop.

Arguments

• objective - Function to minimize.
• n_iter - Number of iterations (function evaluations after initial design).

pub fn optimize_batch<F>( &mut self, objective: F, n_rounds: usize, batch_size: usize, ) -> OptimizeResult<BayesianOptResult>

where F: Fn(&ArrayView1<'_, f64>) -> f64,

Run batch optimization, evaluating batch_size points in parallel per round.

Uses the Kriging Believer strategy: after selecting a candidate, the GP is updated with a fantasised observation at the predicted mean.

pub fn optimize_multi_objective<F>( &mut self, objectives: &[F], n_iter: usize, ) -> OptimizeResult<BayesianOptResult>

where F: Fn(&ArrayView1<'_, f64>) -> f64,

Multi-objective optimization via ParEGO scalarization.

Uses random weight vectors to scalarise the objectives into a single augmented Chebyshev function, then runs standard BO on the scalarization.

Arguments

• objectives - Vector of objective functions to minimize.
• n_iter - Number of sequential iterations.

pub fn ask(&mut self) -> OptimizeResult<Array1<f64>>

Get the ask interface: suggest the next point to evaluate.

pub fn tell(&mut self, x: Array1<f64>, y: f64) -> OptimizeResult<()>

Tell interface: update with an observation.

pub fn best(&self) -> Option<&Observation>

Get the current best observation.

pub fn observations(&self) -> &[Observation]

Get all observations.

pub fn n_observations(&self) -> usize

Number of observations.

pub fn surrogate(&self) -> &GpSurrogate

Get reference to the GP surrogate.