Struct FeedForwardNetwork 

pub struct FeedForwardNetwork<T: Float + Debug + ScalarOperand + Send + Sync + 'static> { /* private fields */ }

Feed-forward network implementation

Implementations

impl<T: Float + Debug + ScalarOperand + Send + Sync + 'static> FeedForwardNetwork<T>

pub fn new( input_dimension: usize, hidden_dimension: usize, activation: ActivationFunction, ) -> Result<Self>

Create new feed-forward network

pub fn new_with_dropout( input_dimension: usize, hidden_dimension: usize, activation: ActivationFunction, dropout_rate: f64, ) -> Result<Self>

Create with custom dropout rate

pub fn forward(&mut self, input: &Array2<T>) -> Result<Array2<T>>

Forward pass through feed-forward network

pub fn forward_with_cache( &mut self, input: &Array2<T>, ) -> Result<(Array2<T>, FeedForwardCache<T>)>

Forward pass that also returns the activations needed for the backward pass. Dropout is bypassed here so that the cached activations match the ones the gradients are computed against.

pub fn backward( &mut self, cache: &FeedForwardCache<T>, grad_output: &Array2<T>, learning_rate: T, ) -> Result<Array2<T>>

Backward pass. Returns dL/d(input) and applies SGD steps to both linear layers.

pub fn parameter_count(&self) -> usize

Get parameter count

pub fn reset(&mut self) -> Result<()>

Reset all parameters

pub fn set_training(&mut self, training: bool)

Set training mode

pub fn get_activation(&self) -> ActivationFunction

Get activation function

pub fn set_activation(&mut self, activation: ActivationFunction)

Set activation function