Struct SparseLinear 

pub struct SparseLinear { /* private fields */ }

Sparse linear layer (also known as sparse fully connected layer)

This layer implements a linear transformation using sparse weight matrices, which can significantly reduce memory usage and computational cost for networks with high sparsity levels.

Implementations

impl SparseLinear

pub fn new( in_features: usize, out_features: usize, sparsity: f32, use_bias: bool, ) -> TorshResult<Self>

Create a new sparse linear layer

pub fn with_config( in_features: usize, out_features: usize, config: SparseLayerConfig, use_bias: bool, ) -> TorshResult<Self>

Create sparse linear layer with configuration

pub fn from_weight(weight: CsrTensor, bias: Option<Tensor>) -> TorshResult<Self>

Create from existing sparse weight matrix

pub fn forward(&self, input: &Tensor) -> TorshResult<Tensor>

Forward pass

pub fn weight(&self) -> &CsrTensor

Get weight matrix

pub fn weight_mut(&mut self) -> &mut CsrTensor

Get mutable weight matrix

pub fn bias(&self) -> Option<&Tensor>

Get bias vector

pub fn bias_mut(&mut self) -> Option<&mut Tensor>

Get mutable bias vector

pub fn in_features(&self) -> usize

Get input features count

pub fn out_features(&self) -> usize

Get output features count

pub fn sparsity(&self) -> f32

Get current sparsity level

pub fn prune_to_sparsity(&mut self, target_sparsity: f32) -> TorshResult<()>

Update sparsity level by pruning weights

pub fn config(&self) -> &SparseLayerConfig

Get configuration

pub fn output_dimensions(&self, input_dims: &[usize]) -> Vec<usize>

Calculate output dimensions for given input dimensions

pub fn memory_stats(&self) -> SparseMemoryStats

Get memory usage statistics

Trait Implementations

impl SparseLayer for SparseLinear

fn forward( &self, input: &dyn SparseTensor, ) -> TorshResult<Box<dyn SparseTensor>>

Forward pass through the layer

fn parameters(&self) -> Vec<&CsrTensor>

Get layer parameters in sparse format

fn parameters_mut(&mut self) -> Vec<&mut CsrTensor>

Get mutable layer parameters

fn layer_type(&self) -> &'static str

Get layer name/type

fn dimensions(&self) -> (Vec<usize>, Vec<usize>)

Get input and output dimensions

fn sparsity_stats(&self) -> SparseStats

Get sparsity statistics

fn train(&mut self, training: bool)

Set training mode

fn training(&self) -> bool

Check if layer is in training mode