Struct KSvd 

pub struct KSvd {
    pub n_atoms: usize,
    pub sparsity: usize,
    pub n_iter: usize,
}

K-SVD dictionary learning algorithm.

Alternates between a sparse coding stage (OMP) and a dictionary update stage (rank-1 SVD update) to learn an overcomplete dictionary adapted to training data.

Reference: Aharon, Elad & Bruckstein (2006).

Fields

n_atoms: usize

Number of dictionary atoms (columns).

sparsity: usize

Target sparsity per signal.

n_iter: usize

Number of training iterations.

Implementations

impl KSvd

pub fn new(n_atoms: usize, sparsity: usize, n_iter: usize) -> Self

Create a new K-SVD learner.

• n_atoms — number of dictionary atoms
• sparsity — maximum number of atoms per signal
• n_iter — number of alternating optimisation iterations

pub fn fit(&self, signals: &[Vec<f64>]) -> Vec<Vec<f64>>

Learn a dictionary from training signals.

• signals — list of training signals (each of the same length d)

Returns a dictionary matrix D of shape d × n_atoms (columns are atoms, stored as Vec<Vecf64> in row-major form, i.e. D[i][j] is row i, atom j).

pub fn encode(&self, dict: &[Vec<f64>], signal: &[f64]) -> Vec<f64>

Encode a single signal using the learned dictionary.

Returns a sparse coefficient vector of length n_atoms.

pub fn reconstruct(dict: &[Vec<f64>], code: &[f64]) -> Vec<f64>

Reconstruct a signal from its sparse code and dictionary.

Returns D c where D is the dictionary and c is the code.