Struct BayesianSearch 

pub struct BayesianSearch { /* private fields */ }

Sequential model-based (Bayesian) parameter optimiser.

Finds near-optimal strategy parameters in a fraction of the evaluations required by exhaustive [GridSearch], making it practical for high-dimensional spaces or continuous float ranges.

Returns the same OptimizationReport as [GridSearch], so the two are drop-in interchangeable and both work with WalkForwardConfig.

Overfitting Warning

Results are in-sample only. Follow up with WalkForwardConfig or a held-out test window to obtain an unbiased out-of-sample estimate.

Implementations

impl BayesianSearch

pub fn new() -> Self

Create a new Bayesian search with no parameters defined yet.

pub fn param(self, name: impl Into<String>, range: ParamRange) -> Self

Add a named parameter range to search over.

Use ParamRange::int_bounds / ParamRange::float_bounds for continuous ranges (recommended) or any ParamRange variant.

pub fn optimize_for(self, metric: OptimizeMetric) -> Self

Set the metric to optimise for (defaults to OptimizeMetric::SharpeRatio).

pub fn max_evaluations(self, n: usize) -> Self

Maximum total strategy evaluations, including the initial LHS phase (default: 100).

pub fn initial_points(self, n: usize) -> Self

Number of initial random (LHS) samples before the surrogate is fitted (default: 10).

Clamped to [2, max_evaluations]. More initial points improve surrogate quality at the cost of fewer sequential refinement steps.

pub fn ucb_beta(self, beta: f64) -> Self

UCB exploration–exploitation coefficient β (default: 2.0).

Higher values drive broader exploration of uncertain regions; lower values concentrate search near already-good parameter sets.

pub fn seed(self, seed: u64) -> Self

PRNG seed for reproducible runs (default: 42).

pub fn run<S, F>( &self, symbol: &str, candles: &[Candle], config: &BacktestConfig, factory: F, ) -> Result<OptimizationReport>

where S: Strategy, F: Fn(&HashMap<String, ParamValue>) -> S,

Run the Bayesian search.

symbol is used only for labelling in the returned results.

factory receives the current parameter map and returns a strategy instance. Parameter sets incompatible with the candle series (warmup too long) are silently skipped.

Returns an error only when no parameters are defined or every evaluation was skipped due to insufficient data.

Trait Implementations

impl Clone for BayesianSearch

fn clone(&self) -> BayesianSearch

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for BayesianSearch

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

Formats the value using the given formatter.

impl Default for BayesianSearch

fn default() -> BayesianSearch

Returns the “default value” for a type.