Struct SgdrScheduler 

pub struct SgdrScheduler {
    pub max_lr: f64,
    pub min_lr: f64,
    pub t_0: usize,
    pub t_mult: f64,
    /* private fields */
}

SGDR: Stochastic Gradient Descent with Warm Restarts scheduler.

Based on “SGDR: Stochastic Gradient Descent with Warm Restarts” (Loshchilov & Hutter, 2017). Periodically resets the learning rate to a high value and then decays it using cosine annealing.

Fields

max_lr: f64

Initial learning rate (after restart).

min_lr: f64

Minimum learning rate.

t_0: usize

Initial period length.

t_mult: f64

Period multiplication factor after each restart.

Implementations

impl SgdrScheduler

pub fn new(max_lr: f64, min_lr: f64, t_0: usize, t_mult: f64) -> Self

Create a new SGDR scheduler.

Arguments

• max_lr - Maximum learning rate (used at restart)
• min_lr - Minimum learning rate
• t_0 - Initial period length
• t_mult - Period multiplication factor (typically 1.0 or 2.0)

Trait Implementations

impl Clone for SgdrScheduler

fn clone(&self) -> SgdrScheduler

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SgdrScheduler

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

Formats the value using the given formatter.

impl LrScheduler for SgdrScheduler

fn step(&mut self, optimizer: &mut dyn Optimizer)

Update learning rate based on current step/epoch.

fn get_lr(&self) -> f64

Get current learning rate.

fn state_dict(&self) -> HashMap<String, f64>

Get scheduler state as a dictionary.

fn load_state_dict(&mut self, state: &HashMap<String, f64>) -> TrainResult<()>

Load scheduler state from a dictionary.