Struct LambdaLr 

pub struct LambdaLr { /* private fields */ }

Lambda learning-rate scheduler.

Computes lr = base_lr * lr_lambda(step_count) at each step, where lr_lambda is a user-provided closure mapping epoch to a multiplicative factor.

Implementations

impl LambdaLr

pub fn new(base_lr: f32, lr_lambda: Box<dyn Fn(usize) -> f32>) -> Self

Creates a lambda scheduler with the given base learning rate and lambda function. The lambda receives the current epoch (after increment) and returns a multiplicative factor applied to base_lr.

pub fn base_lr(&self) -> f32

Returns the base learning rate.

pub fn current_lr(&self) -> f32

Returns the current learning rate.

pub fn step_count(&self) -> usize

Returns the current step count.

pub fn epoch(&self) -> usize

Returns number of already-processed step calls.

pub fn reset(&mut self)

Resets internal state.

pub fn step<O: LearningRate>( &mut self, optimizer: &mut O, ) -> Result<f32, OptimError>

Advances scheduler by one epoch and returns resulting optimizer LR.

Trait Implementations

impl LrScheduler for LambdaLr

fn step<O: LearningRate>( &mut self, optimizer: &mut O, ) -> Result<f32, OptimError>

Advances scheduler by one epoch and returns resulting optimizer LR.

fn epoch(&self) -> usize

Returns number of already-processed step calls.

fn reset(&mut self)

Resets scheduler internal state.