Struct SagaOptimizer 

pub struct SagaOptimizer {
    pub max_epochs: usize,
    pub tol: Float,
    pub learning_rate: Option<Float>,
    pub random_state: Option<u64>,
}

SAGA optimizer (improved SAG with support for non-smooth penalties)

Fields

max_epochs: usize

Maximum number of epochs

tol: Float

Convergence tolerance

learning_rate: Option<Float>

Learning rate

random_state: Option<u64>

Random seed

Implementations

impl SagaOptimizer

pub fn minimize<F, GradF, ProxG>( &self, _f_i: F, grad_f_i: GradF, prox_g: ProxG, x0: Array1<Float>, n_samples: usize, ) -> Result<Array1<Float>, String>

where F: Fn(&Array1<Float>, usize) -> Float, GradF: Fn(&Array1<Float>, usize) -> Array1<Float>, ProxG: Fn(&Array1<Float>, Float) -> Array1<Float>,

Minimize using SAGA for composite objectives: f(x) + g(x) where f is smooth (finite-sum) and g is possibly non-smooth (e.g., L1 penalty)

Trait Implementations

impl Default for SagaOptimizer

fn default() -> Self

Returns the “default value” for a type.