Struct TransformerPerformancePredictor 

pub struct TransformerPerformancePredictor<T: Float + Debug + ScalarOperand + Send + Sync + 'static> { /* private fields */ }

Performance predictor for transformer variants.

Wraps the real predictor from crate::adaptive::predictor: a Xavier-initialized fixed random feature map with ridge-fitted output heads, a bounded prediction cache with real hit accounting, and a predictive standard deviation derived from the ridge posterior.

Before Self::train has been called, Self::predict_improvement answers “no information” (zero means, unit uncertainty, zero confidence) instead of the constants 0.15 / 0.92 / 0.85 / 0.05 it used to fabricate.

Implementations

impl<T: Float + Debug + ScalarOperand + Send + Sync + 'static> TransformerPerformancePredictor<T>

pub fn new(config: &AdaptiveConfig<T>) -> Result<Self>

Build the predictor from the adaptive configuration.

The configuration is genuinely read: prediction_horizon sizes both the hidden feature map and the cache, and adaptation_lr supplies the ridge coefficient (a smaller adaptation rate means a more conservative, more strongly regularized fit).

pub fn train( &mut self, samples: &[PredictorSample], ) -> Result<PredictorFitReport>

Fit the prediction heads on observed (landscape+adaptation, outcome) pairs. Returns the training RMSE of each head.

pub fn is_trained(&self) -> bool

Whether the heads have been fitted.

pub fn cache_hit_rate(&self) -> f64

Observed cache hit rate.

pub fn uncertainty_method(&self) -> UncertaintyMethod

Uncertainty estimation method in use.

pub fn prediction_horizon(&self) -> usize

Prediction horizon this predictor was configured with.

pub fn extract_features( landscape: &LandscapeAnalysis<T>, adaptation: &ArchitectureAdaptation<T>, ) -> PredictionFeatures

Turn a landscape analysis plus a proposed adaptation into the real feature vector the network consumes.

pub fn predict_improvement( &mut self, landscape: &LandscapeAnalysis<T>, adaptation: &ArchitectureAdaptation<T>, ) -> Result<PerformancePrediction<T>>

Trait Implementations

impl<T: Debug + Float + Debug + ScalarOperand + Send + Sync + 'static> Debug for TransformerPerformancePredictor<T>

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

Formats the value using the given formatter.