Struct Dgrdl 

pub struct Dgrdl<T> { /* private fields */ }

Main DGRDL implementation

Implements the Dual Graph Regularized Dictionary Learning algorithm which learns a sparse dictionary representation while preserving local geometric structure in both sample and feature spaces through graph regularization.

Fields

• params - Configuration parameters for the algorithm

Implementations

impl<T: BixverseFloat> Dgrdl<T>

pub fn new(params: DgrdlParams<T>) -> Self

Create a new DGRDL instance

Params

• params - Configuration parameters for the algorithm

Returns

New DGRDL instance ready for training

pub fn fit( &mut self, data: &MatRef<'_, T>, seed: usize, verbose: bool, ) -> DgrdlResults<T>

Run DGRDL algorithm on input data (with initial hyper parameters)

Implements the main optimization loop alternating between sparse coding and dictionary update steps. The algorithm minimizes: ||Y - DX||²_F + α·tr(D^T·D·L_s) + β·tr(X·L_f·X^T) + sparsity constraint where Y is data, D is dictionary, X is coefficients, L_s and L_f are sample and feature Laplacians respectively.

Params

• data - Input data matrix of size n_samples × n_features
• seed - Seed for dictionary initialisation.
• verbose - Whether to print iteration progress

Returns

DgrdlResults containing the learned dictionary, coefficients, and Laplacians

pub fn fit_with_params( &mut self, data: &MatRef<'_, T>, params: DgrdlParams<T>, seed: usize, verbose: bool, ) -> DgrdlResults<T>

Run DGRDL algorithm on input data (with provided hyperparameters)

Implements the main optimization loop alternating between sparse coding and dictionary update steps. The algorithm minimizes: ||Y - DX||²_F + α·tr(D^T·D·L_s) + β·tr(X·L_f·X^T) + sparsity constraint where Y is data, D is dictionary, X is coefficients, L_s and L_f are sample and feature Laplacians respectively.

Params

• data - Input data matrix of size n_samples × n_features
• params - The DgrdlParams containing the parameters you wish to use.
• seed - Seed for dictionary initialisation.
• verbose - Whether to print iteration progress

Returns

DgrdlResults containing the learned dictionary, coefficients, and Laplacians

pub fn grid_search( &mut self, data: &MatRef<'_, T>, dict_sizes: &[usize], k_neighbours_iters: &[usize], seeds: &[usize], verbose: bool, ) -> Vec<DgrdlObjectives<T>>

Run a hyperparameter search over DGRDL

Helper function to sweep over a set of hyperparameters and return different objectives from the algorithm for subsequent visualisation.

Params

• data - Input data matrix of size n_samples × n_features
• dict_sizes - Slice of different dictionary sizes you wish to test for.
• k_neighbours_iters - Slice of different k neighbours to iterate over.
• seeds - Slice of random initialisations to test.
• verbose - Controls verbosity of the function

Returns

DgrdlResults containing the learned dictionary, coefficients, and Laplacians

Trait Implementations

impl<T: Debug> Debug for Dgrdl<T>

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

Formats the value using the given formatter.