Struct TrainingDynamicsAnalyzer 

pub struct TrainingDynamicsAnalyzer { /* private fields */ }

Training dynamics analyzer for monitoring and analyzing training behavior.

Implementations

impl TrainingDynamicsAnalyzer

pub fn new() -> Self

Create a new training dynamics analyzer.

pub fn with_config(config: TrainingAnalysisConfig) -> Self

Create a new analyzer with custom configuration.

pub fn add_metrics(&mut self, metrics: ModelPerformanceMetrics)

Add new training metrics for analysis.

pub fn record_training_dynamics(&mut self, _dynamics: TrainingDynamics)

Record training dynamics information.

pub fn analyze_training_dynamics(&self) -> TrainingDynamics

Analyze current training dynamics.

pub fn detect_convergence_status(&self) -> ConvergenceStatus

Detect current convergence status.

pub fn assess_training_stability(&self) -> TrainingStability

Assess training stability.

pub fn calculate_learning_efficiency(&self) -> f64

Calculate learning efficiency score.

pub fn detect_overfitting_indicators(&self) -> Vec<OverfittingIndicator>

Detect overfitting indicators from the recorded training metrics.

ModelPerformanceMetrics carries no validation split, so the validation-flavoured variants of OverfittingIndicator (TrainValidationGap, ValidationLossIncreasing, HighVarianceInValidation) are never raised here – they exist for callers that really do hold validation data. The two signals this can honestly report are a collapsed training loss and high variance of the training loss.

This previously pushed PerfectTrainingAccuracy whenever the mean training LOSS fell below 0.01 (loss is not accuracy) and HighVarianceInValidation from the variance of the TRAINING loss (there is no validation series to take a variance of).

pub fn detect_underfitting_indicators(&self) -> Vec<UnderfittingIndicator>

Detect underfitting indicators.

pub fn detect_plateau(&self) -> Option<PlateauInfo>

Detect plateau in training.

pub fn generate_training_recommendations(&self) -> Vec<TrainingRecommendation>

Generate training recommendations based on current dynamics.

pub fn clear(&mut self)

Clear analysis history.

pub fn get_training_state(&self) -> &TrainingState

Get current training state information.

pub async fn generate_report(&self) -> Result<TrainingDynamicsReport>

Generate comprehensive training dynamics report.

Trait Implementations

impl Debug for TrainingDynamicsAnalyzer

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

Formats the value using the given formatter.

impl Default for TrainingDynamicsAnalyzer

fn default() -> Self

Returns the “default value” for a type.