Struct SGDOptimizer 

pub struct SGDOptimizer { /* private fields */ }

Stochastic gradient descent optimizer with momentum, Nesterov, and weight decay support.

Implementations

impl SGDOptimizer

pub fn new(config: SGDConfig) -> Self

Create a new optimizer with the given configuration.

pub fn register_parameter(&mut self, name: &str, initial_values: Vec<f64>)

Register a named parameter vector. The velocity buffer is initialised to zeros with the same length as initial_values.

pub fn step( &mut self, gradients: &HashMap<String, Vec<f64>>, ) -> Result<(), String>

Execute one optimiser step.

gradients must contain exactly the same keys as the registered parameters, and each gradient vector must match the corresponding parameter length.

pub fn get_parameter(&self, name: &str) -> Option<&[f64]>

Return a slice of the current parameter values, if the name exists.

pub fn get_velocity(&self, name: &str) -> Option<&[f64]>

Return a slice of the velocity buffer, if the name exists.

pub fn set_learning_rate(&mut self, lr: f64)

Dynamically change the learning rate.

pub fn parameter_count(&self) -> usize

Total number of registered scalar parameter values.

pub fn step_count(&self) -> u64

How many optimiser steps have been performed.

pub fn zero_velocities(&mut self)

Reset all velocity buffers to zero.

pub fn stats(&self) -> SGDOptimizerStats

Return a snapshot of the optimizer statistics.

Trait Implementations

impl Clone for SGDOptimizer

fn clone(&self) -> SGDOptimizer

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SGDOptimizer

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

Formats the value using the given formatter.