Struct OutputProjection 

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

Output projection for generating parameter updates

Implementations

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

pub fn weight_snapshot(&self) -> &Array2<T>

The projection matrix, shaped (output_size, input_size).

pub fn bias_snapshot(&self) -> &Array1<T>

The projection bias, length output_size.

impl<T: Float + Debug + Send + Sync + 'static + Default + Clone> OutputProjection<T>

pub fn new( input_size: usize, output_size: usize, output_transform: OutputTransform, ) -> Result<Self>

Create a new output projection with Xavier-initialized weights.

The output-side transform (tanh / scaling / …) named by output_transform is applied by the caller (see LSTMOptimizer::generate_updates); this projection itself computes the underlying linear map W x + b.

pub fn forward(&self, input: &Array1<T>) -> Result<Array1<T>>

Forward pass through output projection: output = W * input + b.

pub fn reset(&mut self, input_size: usize, output_size: usize)

Re-randomize the projection weights and zero the bias.

Used when the target output dimension changes at runtime (e.g. the flattened parameter count the optimizer is asked to update does not match this projection’s configured output_features): a fresh Xavier draw at the new shape is the honest response, since there is no meaningful way to reuse weights trained for a different output size.

pub fn output_size(&self) -> usize

Current output dimension.

Trait Implementations

impl<T: Clone + Float + Debug + Send + Sync + 'static> Clone for OutputProjection<T>

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T: Debug + Float + Debug + Send + Sync + 'static> Debug for OutputProjection<T>

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

Formats the value using the given formatter.