Enum SchedulerType 

pub enum SchedulerType {
    Constant,
    StepDecay {
        step_size: usize,
        gamma: f32,
    },
    Exponential {
        gamma: f32,
    },
    CosineAnnealing {
        t_max: usize,
        eta_min: f32,
    },
    WarmupLinear {
        warmup_steps: usize,
        total_steps: usize,
    },
    ReduceOnPlateau {
        factor: f32,
        patience: usize,
        min_lr: f32,
    },
}

Learning rate scheduling strategies

Variants

Constant

Constant learning rate throughout training

StepDecay

Step decay: multiply learning rate by gamma every step_size epochs Formula: lr = base_lr * gamma^(epoch / step_size)

Fields

step_size: usize

gamma: f32

Exponential

Exponential decay: multiply learning rate by gamma each epoch Formula: lr = base_lr * gamma^epoch

Fields

gamma: f32

CosineAnnealing

Cosine annealing with warm restarts Formula: lr = eta_min + 0.5 * (base_lr - eta_min) * (1 + cos(pi * (epoch % t_max) / t_max))

Fields

t_max: usize

eta_min: f32

WarmupLinear

Warmup phase followed by linear decay Linearly increases lr from 0 to base_lr over warmup_steps, then linearly decreases to 0 over remaining steps

Fields

warmup_steps: usize

total_steps: usize

ReduceOnPlateau

Reduce learning rate when a metric plateaus Useful for online learning scenarios

Fields

factor: f32

patience: usize

min_lr: f32

Trait Implementations

impl Clone for SchedulerType

fn clone(&self) -> SchedulerType

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SchedulerType

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

Formats the value using the given formatter.