Struct KernelMatrix 

pub struct KernelMatrix { /* private fields */ }

A square kernel matrix (n×n, symmetric positive semi-definite).

All matrix operations are implemented from scratch over Vec<Vec<f64>>.

Implementations

impl KernelMatrix

pub fn new(data: Vec<Vec<f64>>) -> Result<KernelMatrix, AlignmentError>

Construct from a row-major Vec<Vec<f64>>.

Returns Err(AlignmentError::NonSquareMatrix) if any row has a length different from the number of rows.

pub fn from_flat(flat: &[f64], n: usize) -> Result<KernelMatrix, AlignmentError>

Construct from a flat slice of length n*n in row-major order.

pub fn identity(n: usize) -> KernelMatrix

Construct the n×n identity kernel matrix.

pub fn from_labels(labels: &[f64]) -> KernelMatrix

Construct the “ideal” label kernel: K[i,j] = 1.0 if labels[i] == labels[j], else -1.0.

This is equivalent to the outer product of the label sign vector and is the target used in KTA for binary or multi-class classification.

pub fn get(&self, i: usize, j: usize) -> f64

Return element (i, j).

pub fn n(&self) -> usize

Return the dimension n (number of rows/columns).

pub fn trace(&self) -> f64

Compute the trace: sum_i K[i,i].

pub fn frobenius_norm_sq(&self) -> f64

Compute ||K||_F^2 = sum_{i,j} K[i,j]^2.

pub fn frobenius_inner( &self, other: &KernelMatrix, ) -> Result<f64, AlignmentError>

Compute <K1, K2>_F = sum_{i,j} K1[i,j] * K2[i,j].

Returns Err(AlignmentError::DimensionMismatch) if the matrices have different sizes.

pub fn center(&self) -> KernelMatrix

Double-center the kernel matrix: K_c = H * K * H where H = I - (1/n) * 1*1^T is the centering matrix.

Equivalent to:

K_c[i,j] = K[i,j] - row_mean[i] - col_mean[j] + grand_mean

Trait Implementations

impl Clone for KernelMatrix

fn clone(&self) -> KernelMatrix

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for KernelMatrix

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

Formats the value using the given formatter.