Struct Sgd 

pub struct Sgd {
    pub learning_rate: f64,
    pub momentum: f64,
    pub weight_decay: f64,
    pub nesterov: bool,
    /* private fields */
}

Stochastic Gradient Descent with optional momentum, Nesterov lookahead, and L2 weight decay.

Update rule (with momentum): v ← momentum · v + lr · (grad + weight_decay · params) params ← params - v (vanilla momentum) params ← params - (momentum · v + lr · grad) (Nesterov)

Fields

learning_rate: f64

Base learning rate.

momentum: f64

Momentum coefficient (0 = vanilla SGD).

weight_decay: f64

L2 weight-decay coefficient.

nesterov: bool

Use Nesterov momentum instead of classical momentum.

Implementations

impl Sgd

pub fn new(learning_rate: f64, momentum: f64) -> Self

Create a new SGD optimizer.

pub fn step( &mut self, params: &mut Vec<f64>, grad: &[f64], ) -> OptimizeResult<()>

Perform one SGD update step.

Errors

Returns OptimizeError::InvalidInput if params and grad have different lengths.

pub fn zero_velocity(&mut self, n: usize)

Reset velocity buffer to zeros (for new training run).

Trait Implementations

impl Clone for Sgd

fn clone(&self) -> Sgd

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Sgd

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

Formats the value using the given formatter.