Struct NewtonCG

pub struct NewtonCG<L, F> { /* private fields */ }

Newton-Conjugate-Gradient (Newton-CG) method

The Newton-CG method (also called truncated Newton method) uses a modified CG to approximately solve the Newton equations. After a search direction is found, a line search is performed.

Requirements on the optimization problem

The optimization problem is required to implement Gradient and Hessian.

Reference

Jorge Nocedal and Stephen J. Wright (2006). Numerical Optimization. Springer. ISBN 0-387-30303-0.

Implementations

impl<L, F> NewtonCG<L, F>

where F: ArgminFloat,

pub fn new(linesearch: L) -> Self

Construct a new instance of NewtonCG

Example

let ncg: NewtonCG<_, f64> = NewtonCG::new(linesearch);

pub fn with_curvature_threshold(self, threshold: F) -> Self

Set curvature threshold

Defaults to 0.

Example

let ncg: NewtonCG<_, f64> = NewtonCG::new(linesearch).with_curvature_threshold(1e-6);

pub fn with_tolerance(self, tol: F) -> Result<Self, Error>

Set tolerance for the stopping criterion based on cost difference

Must be larger than 0 and defaults to EPSILON.

Example

let ncg: NewtonCG<_, f64> = NewtonCG::new(linesearch).with_tolerance(1e-6)?;

Trait Implementations

impl<L: Clone, F: Clone> Clone for NewtonCG<L, F>

fn clone(&self) -> NewtonCG<L, F>

Returns a copy of the value.

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

Performs copy-assignment from source.

impl<'de, L, F> Deserialize<'de> for NewtonCG<L, F>

where L: Deserialize<'de>, F: Deserialize<'de>,

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl<L, F> Serialize for NewtonCG<L, F>

where L: Serialize, F: Serialize,

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl<O, L, P, G, H, F> Solver<O, IterState<P, G, (), H, (), F>> for NewtonCG<L, F>

where O: Gradient<Param = P, Gradient = G> + Hessian<Param = P, Hessian = H>, P: Clone + ArgminSub<P, P> + ArgminDot<P, F> + ArgminScaledAdd<P, F, P> + ArgminMul<F, P> + ArgminConj + ArgminZeroLike, G: ArgminL2Norm<F> + ArgminMul<F, P>, H: Clone + ArgminDot<P, P>, L: Clone + LineSearch<P, F> + Solver<O, IterState<P, G, (), (), (), F>>, F: ArgminFloat + ArgminL2Norm<F>,

fn name(&self) -> &str

Name of the solver. Mainly used in Observers.

fn next_iter( &mut self, problem: &mut Problem<O>, state: IterState<P, G, (), H, (), F>, ) -> Result<(IterState<P, G, (), H, (), F>, Option<KV>), Error>

Computes a single iteration of the algorithm and has access to the optimization problem definition and the internal state of the solver. Returns an updated state and optionally a KV which holds key-value pairs used in Observers.

fn terminate( &mut self, state: &IterState<P, G, (), H, (), F>, ) -> TerminationStatus

Used to implement stopping criteria, in particular criteria which are not covered by (terminate_internal.

fn init( &mut self, _problem: &mut Problem<O>, state: I, ) -> Result<(I, Option<KV>), Error>

Initializes the algorithm.

fn terminate_internal(&mut self, state: &I) -> TerminationStatus

Checks whether basic termination reasons apply.