Struct AdvancedMLDebugger 

pub struct AdvancedMLDebugger { /* private fields */ }

Advanced ML debugger

Implementations

impl AdvancedMLDebugger

pub fn new(config: AdvancedMLDebuggingConfig) -> Self

Create a new advanced ML debugger

pub async fn analyze_layer_wise_learning_rates( &mut self, layer_gradients: &HashMap<String, ArrayD<f32>>, layer_weights: &HashMap<String, ArrayD<f32>>, current_lr: f64, loss_history: &[f64], ) -> Result<LayerWiseLRAnalysisResult>

Perform layer-wise learning rate analysis

pub async fn analyze_model_sensitivity( &mut self, model_params: &HashMap<String, f64>, performance_metrics: &[f64], architecture_config: &HashMap<String, f64>, ) -> Result<ModelSensitivityAnalysisResult>

Perform comprehensive model sensitivity analysis.

Always returns a structured error. A sensitivity analysis measures how performance responds when a factor is varied, which requires re-evaluating the model at perturbed hyperparameters, architectures, dataset sizes and seeds. This method receives only the CURRENT parameter values plus a flat &[f64] of already-observed metrics – one point per factor – so no response curve, optimal range, stability region or robustness score is derivable from its inputs.

It used to return a fully populated ModelSensitivityAnalysisResult assembled from hardcoded constants: sensitivity_score: 0.8, optimal_range: (0.0001, 0.01), current_size: 10000, most_important_features: ["feature_1", "feature_2"], and a surprising_findings string, none of which looked at model_params, performance_metrics or architecture_config. The five helpers that produced those constants have been deleted.

The result types remain public so a caller that really does run a parameter sweep can construct and share one.

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

Generate comprehensive advanced ML debugging report

Trait Implementations

impl Debug for AdvancedMLDebugger

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

Formats the value using the given formatter.