Struct MetaLearner 

pub struct MetaLearner<A: Float + Send + Sync> { /* private fields */ }

Meta-learning system for streaming optimization

Implementations

impl<A: Float + Default + Clone + Sum + Send + Sync + Debug> MetaLearner<A>

pub fn new(config: &StreamingConfig) -> Result<Self, String>

Creates a new meta-learner

pub fn update_context_signals( &mut self, resource_state: Vec<A>, drift_indicators: Vec<A>, )

Supplies the resource and drift signals the meta-learner cannot observe for itself.

ML5: extract_meta_state used to fill resource_state with the constants [0.5, 0.3] and drift_indicators with [0.1], so the bandit’s context vector was two thirds fabricated and identical on every call. The owning optimizer knows the real values and reports them here.

pub fn update_experience( &mut self, state: MetaState<A>, action: MetaAction<A>, reward: A, ) -> Result<(), String>

Updates the meta-learner with new experience

pub fn register_transfer_source( &mut self, source_id: String, experiences: Vec<MetaExperience<A>>, source_characteristics: Vec<A>, ) -> Result<(), String>

Registers a source domain whose experiences may be replayed into this learner when transfer learning is enabled.

Returns an error when MetaLearningConfig::enable_transfer_learning is off, rather than accepting the source and never using it (CF1: that flag previously had no reader at all, so transfer learning was neither on nor off — it simply did not exist).

pub fn transfer_metrics(&self) -> Option<&TransferMetrics<A>>

Measured transfer-learning outcomes, or None when transfer learning is disabled.

pub fn recommend_adaptations( &mut self, _current_data: &[StreamingDataPoint<A>], performance_tracker: &PerformanceTracker<A>, ) -> Result<Vec<Adaptation<A>>, String>

Recommends adaptations based on current state.

_current_data is accepted for signature stability but is not read: the meta-state the bandit consumes is [performance, resource, drift] with a layout the feature scaler in super::meta_bandit is fitted against, and per-batch data characteristics have no slot in it. The owning optimizer already derives its data statistics separately (compute_data_statistics), and resource/drift signals reach the meta-learner through Self::update_context_signals. Feeding data characteristics into the bandit would require widening MetaState and refitting the scaler, which is a design change rather than a wiring fix.

pub fn apply_adaptation( &mut self, adaptation: &Adaptation<A>, ) -> Result<(), String>

Applies adaptation to meta-learning system

pub fn get_effectiveness_score(&self) -> f32

Gets meta-learning effectiveness score

pub fn get_diagnostics(&self) -> MetaLearningDiagnostics

Gets diagnostic information