Struct PromptOptimizer 

pub struct PromptOptimizer {
    pub variants: Vec<PromptVariant>,
    pub strategy: OptimizationStrategy,
    pub total_trials: u64,
    /* private fields */
}

Tracks prompt variants and selects the best using bandit algorithms.

Fields

variants: Vec<PromptVariant>

All registered variants.

strategy: OptimizationStrategy

Selection strategy.

total_trials: u64

Total number of selection trials performed.

Implementations

impl PromptOptimizer

pub fn new(strategy: OptimizationStrategy) -> Self

Create a new optimizer with the given strategy.

pub fn add_variant(&mut self, template: &str) -> String

Add a variant and return its generated ID.

pub fn select(&mut self, rng_seed: u64) -> Option<&PromptVariant>

Select the next variant to evaluate using the configured strategy.

rng_seed is used for stochastic strategies (EpsilonGreedy, ThompsonSampling). Returns None if no variants are registered.

pub fn record_feedback( &mut self, variant_id: &str, score: f64, cost: f64, tokens_out: u64, )

Record feedback for a variant after evaluation.

pub fn best_variant(&self) -> Option<&PromptVariant>

Return the variant with the highest performance score.

pub fn convergence_ratio(&self) -> f64

Convergence ratio: best_score / 1.0 (max possible).