#include <LineSearchMoreThuente.h>

Static Public Member Functions
template<typename Foo , typename SolverParam >
static void	LineSearch (Foo &f, const SolverParam &param, const Vector &xp, const Vector &drt, const Scalar &step_max, Scalar &step, Scalar &fx, Vector &grad, Scalar &dg, Vector &x)

Detailed Description

template<typename Scalar>
class LBFGSpp::LineSearchMoreThuente< Scalar >

The line search algorithm by Moré and Thuente (1994), currently used for the L-BFGS-B algorithm.

The target of this line search algorithm is to find a step size \(\alpha\) that satisfies the strong Wolfe condition \(f(x+\alpha d) \le f(x) + \alpha\mu g(x)^T d\) and \(|g(x+\alpha d)^T d| \le \eta|g(x)^T d|\). Our implementation is a simplified version of the algorithm in [1]. We assume that \(0<\mu<\eta<1\), while in [1] they do not assume \(\eta>\mu\). As a result, the algorithm in [1] has two stages, but in our implementation we only need the first stage to guarantee the convergence.

Reference: [1] Moré, J. J., & Thuente, D. J. (1994). Line search algorithms with guaranteed sufficient decrease.

Definition at line 26 of file LineSearchMoreThuente.h.

Member Function Documentation

◆ LineSearch()

template<typename Scalar >

template<typename Foo , typename SolverParam >

static void LBFGSpp::LineSearchMoreThuente< Scalar >::LineSearch	(	Foo &	f,
		const SolverParam &	param,
		const Vector &	xp,
		const Vector &	drt,
		const Scalar &	step_max,
		Scalar &	step,
		Scalar &	fx,
		Vector &	grad,
		Scalar &	dg,
		Vector &	x
	)

inlinestatic

Line search by Moré and Thuente (1994).

Parameters

f	A function object such that `f(x, grad)` returns the objective function value at `x`, and overwrites `grad` with the gradient.
param	An `LBFGSParam` or `LBFGSBParam` object that stores the parameters of the solver.
xp	The current point.
drt	The current moving direction.
step_max	The upper bound for the step size that makes x feasible.
step	In: The initial step length. Out: The calculated step length.
fx	In: The objective function value at the current point. Out: The function value at the new point.
grad	In: The current gradient vector. Out: The gradient at the new point.
dg	In: The inner product between drt and grad. Out: The inner product between drt and the new gradient.
x	Out: The new point moved to.

Definition at line 214 of file LineSearchMoreThuente.h.

The documentation for this class was generated from the following file:

LBFGSpp/LineSearchMoreThuente.h

Static Public Member Functions

Detailed Description

Member Function Documentation

◆ LineSearch()