Struct lbfgs::Lbfgs

pub struct Lbfgs { /* fields omitted */ }

LBFGS Buffer

The Limited-memory BFGS algorithm is used to estimate curvature information for the gradient of a function as well as other operators and is often used in numerical optimization and numerical methods in general.

Lbfgs maintains a buffer of pairs (s,y) and values rho (inverse of inner products of s and y)

Methods

impl Lbfgs

pub fn new(problem_size: NonZeroUsize, buffer_size: NonZeroUsize) -> Lbfgs

Create a new L-BFGS instance with a specific problem and L-BFGS buffer size

pub fn with_cbfgs_alpha(self, alpha: f64) -> Self

Update the default C-BFGS alpha

pub fn with_cbfgs_epsilon(self, epsilon: f64) -> Self

Update the default C-BFGS epsilon

pub fn with_sy_epsilon(self, sy_epsilon: f64) -> Self

Update the default sy_epsilon

pub fn reset(&mut self)

"Empties" the buffer

This is a cheap operation as it amount to setting certain internal flags

pub fn apply_hessian(&mut self, g: &mut [f64])

Apply the current Hessian estimate to an input vector

pub fn update_hessian(&mut self, g: &[f64], state: &[f64]) -> UpdateStatus

Saves vectors to update the Hessian estimate

Trait Implementations

impl Debug for Lbfgs

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

Formats the value using the given formatter.