Struct GradAccumulator 

pub struct GradAccumulator {
    pub grad_weights: Vec<f64>,
    pub grad_biases: Vec<f64>,
    pub count: usize,
}

Accumulates gradients from multiple backward passes for mini-batch training.

Fields

grad_weights: Vec<f64>

Accumulated weight gradients.

grad_biases: Vec<f64>

Accumulated bias gradients.

count: usize

Number of samples accumulated.

Implementations

impl GradAccumulator

pub fn new(n_weights: usize, n_biases: usize) -> Self

Create a new accumulator sized for n_weights weights and n_biases biases.

pub fn accumulate(&mut self, gw: &[f64], gb: &[f64])

Add a set of gradients (accumulate without dividing).

pub fn mean_grads(&self) -> (Vec<f64>, Vec<f64>)

Compute mean gradients (divide by count) and return them.

pub fn zero(&mut self)

Zero all accumulated gradients and reset count.

Trait Implementations

impl Clone for GradAccumulator

fn clone(&self) -> GradAccumulator

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GradAccumulator

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

Formats the value using the given formatter.