Struct GpuTrainer 

pub struct GpuTrainer {
    pub net: GpuNeuralNet,
    pub backprop: BackpropGpu,
    pub learning_rate: f64,
    pub optimizer: OptimizerType,
    pub adam: Option<AdamOptimizer>,
}

Combines a network, a backprop context, and an optimiser for training.

Fields

net: GpuNeuralNet

Neural network being trained.

backprop: BackpropGpu

Backpropagation state.

learning_rate: f64

Learning rate.

optimizer: OptimizerType

Which optimiser to use.

adam: Option<AdamOptimizer>

Adam optimiser instance (only active when optimizer == Adam).

Implementations

impl GpuTrainer

pub fn new( net: GpuNeuralNet, learning_rate: f64, optimizer: OptimizerType, ) -> Self

Create a new trainer wrapping net with the given optimiser.

pub fn train_step(&mut self, input: &[f64], target: &[f64]) -> f64

Execute one training step: forward pass → loss → backward → update.

• input — network input.
• target — ground-truth target.

Returns the MSE loss before the update.

Trait Implementations

impl Debug for GpuTrainer

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

Formats the value using the given formatter.