Struct AdaptiveTuner 

pub struct AdaptiveTuner<A: Float> { /* private fields */ }

Adaptive tuner for dynamic optimization

Implementations

impl<A: Float + Send + Sync> AdaptiveTuner<A>

pub fn new() -> Self

Create a new adaptive tuner.

The default strategy is greedy hill climbing, which needs no search-space budget beyond the parameters themselves; register those with AdaptiveTuner::add_parameter before calling AdaptiveTuner::tune.

pub fn random_seed(&self) -> u64

Seed the stochastic searches draw from.

pub fn set_random_seed(&mut self, seed: u64)

Set the seed the stochastic searches draw from.

Two AdaptiveTuner::tune calls with the same seed, search space and evaluator produce the same sequence of candidates.

pub fn performance_target(&self) -> A

Performance the search stops at once reached.

pub fn set_performance_target(&mut self, target: A)

Set the performance target.

pub fn strategy(&self) -> &TuningStrategy

Search strategy in use.

pub fn set_strategy(&mut self, strategy: TuningStrategy)

Select the search strategy.

pub fn add_parameter(&mut self, parameter: TunableParameter<A>) -> Result<()>

Register a parameter the search may move.

pub fn parameters(&self) -> &[TunableParameter<A>]

Parameters the search may move.

pub fn current_params(&self) -> &HashMap<String, A>

Best candidate found so far.

pub fn best_performance(&self) -> Option<A>

Performance measured for AdaptiveTuner::current_params.

pub fn tuning_history(&self) -> &[TuningRecord<A>]

Every measurement taken so far, oldest first, bounded to the most recent MAX_TUNING_HISTORY entries.

pub fn total_evaluations(&self) -> usize

Total candidates evaluated over the tuner’s lifetime.

pub fn max_evaluations(&self) -> usize

Evaluation budget for a single AdaptiveTuner::tune call.

pub fn set_max_evaluations(&mut self, max_evaluations: usize)

Set the evaluation budget for a single AdaptiveTuner::tune call.

pub fn tune<F>(&mut self, evaluate: F) -> Result<TuningOutcome<A>>

where F: FnMut(&HashMap<String, A>) -> Result<TuningObservation<A>>,

Run the search behind the configured TuningStrategy.

evaluate measures one candidate parameter set; it is called at most AdaptiveTuner::max_evaluations times. The search maximizes TuningObservation::performance and stops early once a candidate reaches AdaptiveTuner::performance_target.

The previous best candidate is used as the starting point for greedy search, but the recorded best performance is reset first: the objective a caller measures now (a different workload, a different platform) is not comparable with one measured before, and keeping the old number would silently veto every new measurement.

pub fn cheapest_candidate_meeting_target(&self) -> Option<&TuningRecord<A>>

Best resource usage recorded among the candidates that met the performance target, if any.

Surfaces TuningRecord::resource_usage, which is otherwise only reachable by walking the history: the cheapest candidate that still hits the target is usually the one a caller wants to deploy.

Trait Implementations

impl<A: Debug + Float> Debug for AdaptiveTuner<A>

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

Formats the value using the given formatter.

impl<A: Float + Send + Sync> Default for AdaptiveTuner<A>

fn default() -> Self

Returns the “default value” for a type.