Struct LinearDistiller 

pub struct LinearDistiller { /* private fields */ }

Simple linear distillation model (for testing/simple cases)

Implementations

impl LinearDistiller

pub fn new(n_features: usize, n_classes: usize) -> LinearDistiller

Create a new linear distiller

Arguments

• n_features - Number of input features
• n_classes - Number of output classes

pub fn with_config( n_features: usize, n_classes: usize, config: DistillationConfig, ) -> LinearDistiller

Create with custom config

pub fn forward(&self, features: &[f64]) -> Result<Vec<f64>, AprenderError>

Compute student logits for a single sample

pub fn train_step( &mut self, features: &[f64], teacher_logits: &[f64], hard_labels: &[f64], ) -> Result<f64, AprenderError>

Train on a single sample using teacher’s soft targets

Arguments

• features - Input features
• teacher_logits - Teacher’s output logits
• hard_labels - True one-hot labels

Returns

Loss value before update

pub fn weights(&self) -> &[f64]

Get weights

pub fn biases(&self) -> &[f64]

Get biases

pub fn predict_proba(&self, features: &[f64]) -> Result<Vec<f64>, AprenderError>

Predict class probabilities

pub fn predict(&self, features: &[f64]) -> Result<usize, AprenderError>

Predict class label

Trait Implementations

impl Clone for LinearDistiller

fn clone(&self) -> LinearDistiller

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for LinearDistiller

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

Formats the value using the given formatter.