Struct BackpropGpu 

pub struct BackpropGpu {
    pub gradients: Vec<Vec<f64>>,
}

Backpropagation state: stores per-layer gradient tensors.

Fields

gradients: Vec<Vec<f64>>

Per-layer gradient vectors (same shape as the layer’s weight vector).

Implementations

impl BackpropGpu

pub fn new(net: &GpuNeuralNet) -> Self

Initialise gradient buffers matching the network’s weight shapes.

pub fn backward_pass(&mut self, loss_grad: &[f64])

Perform a mock backward pass given the output loss gradient.

This is a simplified implementation: each layer’s gradient is set to loss_grad[0] times the layer’s weight magnitudes (a stand-in for a real backprop chain rule).

Trait Implementations

impl Clone for BackpropGpu

fn clone(&self) -> BackpropGpu

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BackpropGpu

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

Formats the value using the given formatter.