Struct GradientChecker 

pub struct GradientChecker<T: FloatBounds + ScalarOperand + ToPrimitive> { /* private fields */ }

Gradient checker for neural networks

Implementations

impl<T: FloatBounds + ScalarOperand + ToPrimitive> GradientChecker<T>

pub fn new(config: GradientCheckConfig<T>) -> Self

Create a new gradient checker

pub fn check_network_gradients( &self, network: &mut SimpleMLP<T>, inputs: &Array2<T>, targets: &Array2<T>, loss_fn: &dyn LossFunction<T>, ) -> Result<GradientCheckResults<T>, SklearsError>

Check gradients for a neural network

pub fn check_layer_gradients( &self, layer: &mut DenseLayer<T>, inputs: &Array2<T>, output_gradients: &Array2<T>, ) -> Result<GradientCheckResults<T>, SklearsError>

Check gradients for a single layer

impl<T: FloatBounds + ScalarOperand + ToPrimitive> GradientChecker<T>

Utility functions for gradient checking

pub fn gradients_are_equal(&self, analytical: T, numerical: T) -> bool

Check if gradients are approximately equal

pub fn compute_relative_error(&self, analytical: T, numerical: T) -> T

Compute relative error between two gradients

pub fn generate_report(&self, results: &GradientCheckResults<T>) -> String

Generate a summary report of gradient checking results

Trait Implementations

impl<T: Debug + FloatBounds + ScalarOperand + ToPrimitive> Debug for GradientChecker<T>

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

Formats the value using the given formatter.