Struct zenu_layer::layers::linear::Linear

pub struct Linear<T: Num> { /* private fields */ }

A linear layer in a neural network.

Implementations

impl<T: Num> Linear<T>

pub fn new(in_dim: usize, out_dim: usize) -> Self

Creates a new linear layer with the specified input and output dimensions.

Arguments

• in_dim - The input dimension of the layer.
• out_dim - The output dimension of the layer.

Returns

A new Linear instance.

Trait Implementations

impl<T: Num> Layer<T> for Linear<T>

fn init_parameters(&mut self, seed: Option<u64>)

where StandardNormal: Distribution<T>,

Initializes the layer parameters (weight and bias) randomly or with a specified seed.

Arguments

• seed - An optional seed value for reproducible parameter initialization.

fn parameters(&self) -> Vec<Variable<T>>

Returns a vector of the layer parameters as Variables.

Returns

A vector containing the weight and bias variables of the layer.

fn load_parameters(&mut self, parameters: &[Variable<T>])

Loads pre-trained parameters into the layer.

Arguments

• parameters - A slice of Variables containing the weight and bias parameters.

Panics

Panics if the number of parameters is not 2, or if the shapes of the parameters are invalid.

fn call(&self, input: Variable<T>) -> Variable<T>

Performs a forward pass through the layer, applying the linear transformation to the input.

Arguments

• input - The input data as a Variable.

Returns

The output of the layer as a new Variable.

Panics

Panics if the weight or bias parameters are not initialized.

fn shape_check(&self, input: &Variable<T>)

Checks the shape of the input data to ensure compatibility with the layer.

Arguments

• input - The input data as a reference to a Variable.

Panics

Panics if the input shape is not 2D or if the input dimension does not match the layer’s input dimension.

fn clear_gradients(&self)