Struct CSRTensor 

pub struct CSRTensor {
    pub values: Vec<f32>,
    pub col_indices: Vec<usize>,
    pub row_ptrs: Vec<usize>,
    pub nrows: usize,
    pub ncols: usize,
}

Compressed Sparse Row (CSR) representation.

Memory Layout

For a matrix with nnz non-zeros and nrows rows:

• values: [nnz] - Non-zero values in row-major order
• col_indices: [nnz] - Column index for each value
• row_ptrs: [nrows + 1] - Start index in values/col_indices for each row

Example

Dense:     [[1, 0, 2],    CSR:
            [0, 0, 3],    values: [1, 2, 3, 4]
            [4, 0, 0]]    col_indices: [0, 2, 2, 0]
                          row_ptrs: [0, 2, 3, 4]

Fields

values: Vec<f32>

Non-zero values.

col_indices: Vec<usize>

Column indices for each value.

row_ptrs: Vec<usize>

Row pointers (start index for each row).

nrows: usize

Number of rows.

ncols: usize

Number of columns.

Implementations

impl CSRTensor

pub fn new( values: Vec<f32>, col_indices: Vec<usize>, row_ptrs: Vec<usize>, nrows: usize, ncols: usize, ) -> Result<Self, PruningError>

Create a new CSR tensor from components.

Arguments

• values - Non-zero values
• col_indices - Column index for each value
• row_ptrs - Row pointer array
• nrows - Number of rows
• ncols - Number of columns

pub fn from_dense(tensor: &Tensor) -> Result<Self, PruningError>

Create CSR tensor from dense tensor.

pub fn to_dense(&self) -> Tensor

Convert CSR tensor back to dense tensor.

pub fn nnz(&self) -> usize

Get number of non-zero elements.

pub fn sparsity(&self) -> f32

Get sparsity ratio (fraction of zeros).

pub fn shape(&self) -> [usize; 2]

Get shape as [rows, cols].

pub fn get(&self, row: usize, col: usize) -> f32

Get value at (row, col), or 0 if not present.

pub fn matvec(&self, x: &[f32]) -> Result<Vec<f32>, PruningError>

Sparse matrix-vector multiplication: y = A * x

Arguments

• x - Input vector of length ncols

Returns

Output vector of length nrows

pub fn memory_bytes(&self) -> usize

Memory usage in bytes (approximate).

pub fn dense_memory_bytes(&self) -> usize

Dense memory usage for comparison.

pub fn memory_savings_ratio(&self) -> f32

Memory savings ratio (dense / sparse).

Trait Implementations

impl Clone for CSRTensor

fn clone(&self) -> CSRTensor

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CSRTensor

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

Formats the value using the given formatter.