Struct DenseLayer 

pub struct DenseLayer {
    pub weights: Vec<Vec<f64>>,
    pub biases: Vec<f64>,
    pub activation: Activation,
}

An affine-plus-activation layer y = activation(Wx + b).

Weights are stored row-major as weights[out][in]. The layer itself has no learnable state beyond weights and biases; gradient and optimiser logic lives with the caller (see crate::deep_kernel::gradient).

Fields

weights: Vec<Vec<f64>>

Row-major weight matrix with shape [output_dim][input_dim].

biases: Vec<f64>

Bias vector with length output_dim.

activation: Activation

Element-wise activation applied to the affine pre-activation.

Implementations

impl DenseLayer

pub fn new( weights: Vec<Vec<f64>>, biases: Vec<f64>, activation: Activation, ) -> Result<Self>

Build a layer from raw weights and biases.

Fails when weight rows disagree in width, when the bias vector does not match the weight row count, or when any entry is non-finite.

pub fn input_dim(&self) -> usize

Input dimension (width of each weight row).

pub fn output_dim(&self) -> usize

Output dimension (number of weight rows).

pub fn activation(&self) -> Activation

Activation attached to this layer.

pub fn forward(&self, input: &[f64]) -> Result<Vec<f64>>

Forward pass returning only the post-activation output.

pub fn forward_with_preactivation( &self, input: &[f64], ) -> Result<(Vec<f64>, Vec<f64>)>

Forward pass returning (pre_activation, post_activation).

The pre-activation is retained so callers implementing analytical gradients can feed it back into Activation::derivative.

pub fn parameter_count(&self) -> usize

Count of trainable scalar parameters (weights + biases).

Trait Implementations

impl Clone for DenseLayer

fn clone(&self) -> DenseLayer

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DenseLayer

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

Formats the value using the given formatter.