Struct HyperparameterTuner 

pub struct HyperparameterTuner {
    pub config: TunerConfig,
    pub history: Vec<TuningResult>,
    pub next_trial_id: u64,
}

Hyperparameter tuner supporting random search, grid search, and simplified Bayesian optimization with UCB acquisition.

Fields

config: TunerConfig

history: Vec<TuningResult>

next_trial_id: u64

Implementations

impl HyperparameterTuner

pub fn new(config: TunerConfig) -> Self

Create a new tuner with the given configuration.

pub fn add_spec(&mut self, spec: HpSpec) -> &mut Self

Add a hyperparameter specification to the tuner (builder-style).

pub fn sample_value(spec: &HpSpec, rng: &mut u64) -> HpValue

Sample a single value for one spec using the xorshift64 PRNG.

pub fn sample_config(&self, rng: &mut u64) -> HpConfig

Sample a complete configuration (one value per spec).

pub fn record_result(&mut self, config: HpConfig, score: f64, now: u64) -> u64

Record an evaluation result. Returns the assigned trial_id.

pub fn best_config(&self) -> Option<&TuningResult>

Return a reference to the best trial in history, or None if empty.

pub fn suggest_next(&self, rng: &mut u64) -> HpConfig

Suggest the next configuration to evaluate.

• For RandomSearch / no strategy context: sample a random config.
• For Bayesian: draw 10 random candidates, evaluate UCB for each, and return the one with the highest UCB.

pub fn run_random_search( &mut self, n_trials: u32, scorer: impl FnMut(&HpConfig) -> f64, rng: &mut u64, ) -> Vec<TuningResult>

Run n_trials random evaluations using the provided scorer, record all results, and return them sorted by score (best first).

pub fn grid_configs(&self) -> Vec<HpConfig>

Enumerate all grid configurations.

• Continuous → 5 evenly-spaced values across [lo, hi]
• Discrete → every integer in [lo, hi]
• Categorical → every choice

pub fn run_grid_search( &mut self, scorer: impl FnMut(&HpConfig) -> f64, ) -> Vec<TuningResult>

Enumerate all grid configurations, evaluate each with scorer, record results, and return them sorted best-first.

pub fn run_bayesian( &mut self, n_trials: u32, n_initial: u32, exploration_weight: f64, scorer: impl FnMut(&HpConfig) -> f64, rng: &mut u64, ) -> Vec<TuningResult>

Run Bayesian (UCB) optimization. Draws n_initial random configs to warm-start, then proposes n_trials - n_initial UCB-guided candidates.

pub fn importance_scores(&self) -> HashMap<String, f64>

Compute a simple variance-of-scores importance score for each parameter.

For each parameter, group history trials into equal-width buckets across the parameter’s range and compute the variance of per-bucket mean scores. Returns 0.0 for any parameter if there are fewer than 2 trials.

pub fn stats(&self) -> TunerStats

Aggregate statistics over all recorded trials.

Trait Implementations

impl Debug for HyperparameterTuner

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

Formats the value using the given formatter.