Struct StructuredSparsity 

pub struct StructuredSparsity<A: Float + ScalarOperand + Debug> { /* private fields */ }

Structured sparsity regularizer

Applies Group Lasso regularization with groups defined by a structural pattern (columns, rows, or blocks). This is useful for encouraging structured sparsity in weight matrices, e.g., pruning entire neurons (row sparsity) or features (column sparsity).

Examples

use scirs2_core::ndarray::Array1;
use optirs_core::regularizers::{StructuredSparsity, SparsityPattern, Regularizer};

// Create column-wise structured sparsity for a 3x4 matrix (stored as 12-element vector)
let regularizer = StructuredSparsity::new(0.1_f64, SparsityPattern::Column { num_columns: 4 });

let params = Array1::from_vec(vec![1.0, 0.0, 3.0, 0.0, 5.0, 0.0, 7.0, 0.0, 9.0, 0.0, 11.0, 0.0]);
let penalty = regularizer.penalty(&params).expect("penalty computation failed");

Implementations

impl<A: Float + ScalarOperand + Debug> StructuredSparsity<A>

pub fn new(lambda: A, pattern: SparsityPattern) -> Self

Create a new structured sparsity regularizer

Arguments

• lambda - Regularization strength
• pattern - The structural pattern defining how parameters are grouped

pub fn lambda(&self) -> A

Get the regularization strength

pub fn pattern(&self) -> &SparsityPattern

Get the sparsity pattern

Trait Implementations

impl<A: Clone + Float + ScalarOperand + Debug> Clone for StructuredSparsity<A>

fn clone(&self) -> StructuredSparsity<A>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<A: Debug + Float + ScalarOperand + Debug> Debug for StructuredSparsity<A>

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

Formats the value using the given formatter.

impl<A, D> Regularizer<A, D> for StructuredSparsity<A>

where A: Float + ScalarOperand + Debug, D: Dimension,

fn apply(&self, params: &Array<A, D>, gradients: &mut Array<A, D>) -> Result<A>

Apply regularization to parameters and gradients

fn penalty(&self, params: &Array<A, D>) -> Result<A>

Compute the regularization penalty value