Struct rusty_machine::learning::nnet::net_layer::Linear

pub struct Linear { /* private fields */ }

Linear network layer

Represents a fully connected layer with optional bias term

The parameters are a matrix of weights of size I x N where N is the dimensionality of the output and I the dimensionality of the input

Implementations

impl Linear

pub fn new(input_size: usize, output_size: usize) -> Linear

Construct a new Linear layer

pub fn without_bias(input_size: usize, output_size: usize) -> Linear

Construct a Linear layer without a bias term

Trait Implementations

impl Clone for Linear

fn clone(&self) -> Linear

Returns a copy of the value.

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

Performs copy-assignment from source.

impl Debug for Linear

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

Formats the value using the given formatter.

impl NetLayer for Linear

fn forward(
    &self,
    input: &Matrix<f64>,
    params: MatrixSlice<'_, f64>
) -> LearningResult<Matrix<f64>>

Computes a matrix product

input should have dimensions N x I where N is the number of samples and I is the dimensionality of the input

fn default_params(&self) -> Vec<f64>

Initializes weights using Xavier initialization

weights drawn from gaussian distribution with 0 mean and variance 2/(input_size+output_size)

fn back_input(
    &self,
    out_grad: &Matrix<f64>,
    _: &Matrix<f64>,
    _: &Matrix<f64>,
    params: MatrixSlice<'_, f64>
) -> Matrix<f64>

The gradient of the output of this layer with respect to its input

fn back_params(
    &self,
    out_grad: &Matrix<f64>,
    input: &Matrix<f64>,
    _: &Matrix<f64>,
    _: MatrixSlice<'_, f64>
) -> Matrix<f64>

The gradient of the output of this layer with respect to its parameters

fn param_shape(&self) -> (usize, usize)

The shape of the parameters used by this layer

fn num_params(&self) -> usize

The number of parameters used by this layer

impl Copy for Linear