Struct LearningRateScheduler 

pub struct LearningRateScheduler {
    pub strategy: SchedulerStrategy,
    pub state: LrSchedulerState,
    /* private fields */
}

Multi-strategy learning rate scheduler.

Supports Constant, StepDecay, ExponentialDecay, CosineAnnealing, WarmupCosine, CyclicLR, and ReduceOnPlateau.

Fields

strategy: SchedulerStrategy

The scheduling strategy in use.

state: LrSchedulerState

Mutable runtime state.

Implementations

impl LearningRateScheduler

pub fn new(strategy: SchedulerStrategy) -> Self

Create a new scheduler. The initial learning rate is derived from the strategy variant’s initial_lr (or lr for Constant).

pub fn step(&mut self, epoch: u64) -> f64

Compute the learning rate for the given epoch without mutating significant state (except appending to history and updating current_epoch / current_lr).

For ReduceOnPlateau this ignores plateau tracking — use step_with_loss instead.

pub fn step_with_loss(&mut self, epoch: u64, loss: f64) -> f64

Advance one step using a loss value. For ReduceOnPlateau this updates the internal plateau counter and reduces LR as needed. For other strategies it behaves identically to step.

pub fn current_lr(&self) -> f64

Return the most recently computed learning rate.

pub fn reset(&mut self)

Reset internal state to epoch 0 and recompute the initial LR.

pub fn history(&self) -> &[LrHistory]

Return a slice of the LR history (capped at the last 1 000 entries).

pub fn stats(&self) -> LrStats

Compute aggregate statistics over the recorded history.

pub fn warmup_factor(epoch: u64, warmup_epochs: u64) -> f64

Linear interpolation factor for a warmup phase.

Returns values in [0.0, 1.0] linearly from 0 to 1 as epoch goes from 0 to warmup_epochs. Returns 1.0 when epoch >= warmup_epochs.

pub fn cosine_factor(epoch: u64, t_max: u64) -> f64

Cosine annealing factor: (1 + cos(π * epoch / t_max)) / 2.

Returns values in [0.0, 1.0]. Clamps epoch at t_max.

Trait Implementations

impl Clone for LearningRateScheduler

fn clone(&self) -> LearningRateScheduler

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LearningRateScheduler

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

Formats the value using the given formatter.