Struct SparseSign 

pub struct SparseSign {
    pub rows: usize,
    pub cols: usize,
    pub sparsity: usize,
    /* private fields */
}

Sparse sign random matrix: each column has exactly s non-zeros, each ±1.

This provides an efficient O(s·n) matrix-vector product compared to O(m·n) for dense random matrices. Typically s = O(log(m)) suffices for good embedding properties.

Fields

rows: usize

Number of rows (sketch dimension)

cols: usize

Number of columns (original dimension)

sparsity: usize

Sparsity: number of non-zeros per column

Implementations

impl SparseSign

pub fn new( rows: usize, cols: usize, sparsity: Option<usize>, seed: Option<u64>, ) -> LinalgResult<Self>

Construct a sparse sign matrix with given dimensions and sparsity.

Arguments

• rows - Sketch dimension (number of rows)
• cols - Input dimension (number of columns)
• sparsity - Non-zeros per column (default: max(1, ceil(log2(rows))))
• seed - Optional RNG seed for reproducibility

pub fn apply(&self, x: &ArrayView1<'_, f64>) -> LinalgResult<Array1<f64>>

Apply the sparse sign matrix: y = S * x, result is rows-vector.

Arguments

• x - Input vector of length self.cols

pub fn apply_matrix(&self, a: &ArrayView2<'_, f64>) -> LinalgResult<Array2<f64>>

Apply S to each column of matrix A (result: rows × n).

Computes S * A for a cols × n matrix A.

pub fn apply_transpose( &self, x: &ArrayView1<'_, f64>, ) -> LinalgResult<Array1<f64>>

Apply S^T (transpose): y = S^T * x, result is cols-vector.

Trait Implementations

impl Clone for SparseSign

fn clone(&self) -> SparseSign

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SparseSign

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

Formats the value using the given formatter.