Struct Exponential 

pub struct Exponential { /* private fields */ }

Exponential is a builder for Exponential Learning rate scheduler which allows for more flexible and readable construction.

It implements an exponential decay learning rate scheduler which continuously decreases the learning rate after each epoch. This scheduler is useful for smoothing the training process by gradually decreasing the step size of updates to the model’s parameters, allowing for more precise convergence as training progresses. The rate of decay per epoch is controlled by decay_rate raised to the power of the product of the epoch number and decay_factor.

Implementations

impl Exponential

pub fn initial_lr(self, initial_lr: f32) -> Self

Sets the initial learning rate. This is the starting learning rate before any decay is applied.

Parameters

• initial_lr: The initial learning rate.

pub fn decay_rate(self, decay_rate: f32) -> Self

Sets the decay rate. This is the base for the exponential decay.

Parameters

• decay_rate: The decay rate.

pub fn decay_factor(self, decay_factor: f32) -> Self

Sets the decay factor. This controls how quickly the learning rate decreases. A smaller decay factor results in slower decay.

Parameters

• decay_factor: The decay factor.

pub fn build(self) -> Result<Box<dyn LearningRateScheduler>, NetworkError>

Builds the ExponentialLRScheduler if all required fields are set.

Trait Implementations

impl Default for Exponential

fn default() -> Self

Creates a new ExponentialLRSchedulerBuilder with default values. Default values:

• initial_lr: 0.01
• decay_rate: 0.95
• decay_factor: 0.1