org.ddogleg.optimization.impl

Class TrustRegionLeastSquares

java.lang.Object

org.ddogleg.optimization.impl.TrustRegionLeastSquares

public class TrustRegionLeastSquares
extends java.lang.Object

Trust region based optimization algorithms switch between the optimal Gauss Newton step and gradient descent. A circular region is defined around the current estimate for which the Taylor series expansion is trusted. If the Gauss Newton step falls within this "trust region" then it is used, otherwise the optimal quadratic solution is found subject to the trust region's constraint.

There are several different ways for which to estimate the solution to the trust region problem. The Levenberg-Marquardt algorithm is actually a type of trust region algorithm, even though it was originally proposed before the trust region approach had been.

Constructor Summary

Constructors

Constructor and Description
TrustRegionLeastSquares(double maxRadius, TrustRegionStep stepAlg)

Method Summary

Modifier and Type	Method and Description
double	getError()
double[]	getParameters()
void	initialize(double[] initial)
boolean	isConverged()
boolean	isUpdated()
boolean	iterate() Performs a single iteration.
void	setConvergence(double ftol, double gtol) Specify convergence tolerances
void	setFunction(CoupledJacobian function)

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

TrustRegionLeastSquares

public TrustRegionLeastSquares(double maxRadius,
                               TrustRegionStep stepAlg)

Method Detail

setFunction

public void setFunction(CoupledJacobian function)

setConvergence

public void setConvergence(double ftol,
                           double gtol)

Specify convergence tolerances

Parameters:

gtol - absolute convergence tolerance based on gradient norm. 0 <= gtol

initialize

public void initialize(double[] initial)

iterate

public boolean iterate()

Performs a single iteration.

Returns:

true if it has converged and false if it has not.

getParameters

public double[] getParameters()

getError

public double getError()

isConverged

public boolean isConverged()

isUpdated

public boolean isUpdated()