Struct MatrixScaler 

pub struct MatrixScaler { /* private fields */ }

Matrix Scaling - full affine transformation for maximum calibration flexibility Uses full weight matrix W and bias vector b: calibrated = softmax(W @ logits + b) More expressive than vector scaling but requires more data to avoid overfitting

Implementations

impl MatrixScaler

pub fn new() -> Self

Create a new matrix scaler

pub fn with_regularization(regularization: f64) -> Self

Create matrix scaler with custom regularization

pub fn fit( &mut self, logits: &Array2<f64>, labels: &Array1<usize>, ) -> Result<()>

Fit the matrix scaler to logits and true labels Uses gradient descent with L2 regularization on off-diagonal weights

pub fn transform(&self, logits: &Array2<f64>) -> Result<Array2<f64>>

Transform logits to calibrated probabilities using matrix scaling

pub fn fit_transform( &mut self, logits: &Array2<f64>, labels: &Array1<usize>, ) -> Result<Array2<f64>>

Fit and transform in one step

pub fn parameters(&self) -> Option<(Array2<f64>, Array1<f64>)>

Get the fitted parameters (weight_matrix, bias_vector)

pub fn condition_number(&self) -> Option<f64>

Get the condition number of the weight matrix (for diagnostics) Higher values indicate potential numerical instability

Trait Implementations

impl Clone for MatrixScaler

fn clone(&self) -> MatrixScaler

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MatrixScaler

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

Formats the value using the given formatter.

impl Default for MatrixScaler

fn default() -> Self

Returns the “default value” for a type.