Struct SlbfgsOptimizer 

pub struct SlbfgsOptimizer {
    pub config: SlbfgsConfig,
}

Stochastic L-BFGS optimizer with optional SVRG variance reduction.

Fields

config: SlbfgsConfig

Algorithm configuration.

Implementations

impl SlbfgsOptimizer

pub fn new(config: SlbfgsConfig) -> Self

Create with given configuration.

pub fn default_config() -> Self

Create with default configuration.

pub fn minimize( &self, stoch_f_and_g: &dyn Fn(&[f64], &[usize]) -> (f64, Vec<f64>), full_grad_fn: &dyn Fn(&[f64]) -> (f64, Vec<f64>), n_samples: usize, x0: &[f64], ) -> Result<OptResult, OptimizeError>

Minimize a stochastic objective function using S-L-BFGS.

Arguments

• stoch_f_and_g — stochastic oracle: given x and a batch of indices (subset of 0..n_samples), returns (f, ∇f) on that mini-batch.
• full_grad_fn — deterministic oracle for full gradient at snapshot: (x) → (f, ∇f) over the full dataset.
• n_samples — total number of data points.
• x0 — initial point.

Returns

An OptResult describing the found minimizer and convergence status.