org.ddogleg.optimization

Class FactoryOptimization

java.lang.Object

org.ddogleg.optimization.FactoryOptimization

public class FactoryOptimization
extends java.lang.Object

Creates optimization algorithms using easy to use interfaces. These implementations/interfaces are designed to be easy to use and effective for most tasks. If more control is needed then create an implementation directly.

Constructor Summary

Constructors

Constructor and Description
FactoryOptimization()

Method Summary

Modifier and Type	Method and Description
static QuasiNewtonBFGS_to_UnconstrainedMinimization	createBfgsWithMore94() Returns an implementation of QuasiNewtonBFGS with LineSearchMore94 for the internal line search.
static UnconstrainedLeastSquares	leastSquareLevenberg(double dampInit) Unconstrained least squares Levenberg optimizer for dense problems.
static UnconstrainedLeastSquares	leastSquaresLM(double dampInit, boolean robust) Unconstrained least squares Levenberg-Marquardt (LM) optimizer for dense problems.
static UnconstrainedLeastSquares	leastSquaresTrustRegion(double regionSize, RegionStepType type, boolean robustSolver) Creates a trust region based optimization algorithm for least squares problem.
static UnconstrainedMinimization	unconstrained() Creates a solver for the unconstrained minimization problem.

Methods inherited from class java.lang.Object

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

Constructor Detail

FactoryOptimization

public FactoryOptimization()

Method Detail

unconstrained

public static UnconstrainedMinimization unconstrained()

Creates a solver for the unconstrained minimization problem. Here a function has N parameters and a single output. The goal is the minimize the output given the function and its derivative.

Returns:

UnconstrainedMinimization

createBfgsWithMore94

public static QuasiNewtonBFGS_to_UnconstrainedMinimization createBfgsWithMore94()

Returns an implementation of QuasiNewtonBFGS with LineSearchMore94 for the internal line search. This is a specific implementation of UnconstrainedMinimization and allows full access to all tuning parameters.

Returns:

UnconstrainedMinimization

leastSquaresLM

public static UnconstrainedLeastSquares leastSquaresLM(double dampInit,
                                                       boolean robust)

Unconstrained least squares Levenberg-Marquardt (LM) optimizer for dense problems. There are many different variants of LM and this function provides an easy to use interface for selecting and configuring them. Scaling of function parameters and output might be needed to ensure good results.

Parameters:

dampInit - Initial value of dampening parameter. Tune. Start at around 1e-3.

robust - If true a slower, more robust algorithm that can handle more degenerate cases will be used.

Returns:

UnconstrainedLeastSquares

leastSquareLevenberg

public static UnconstrainedLeastSquares leastSquareLevenberg(double dampInit)

Unconstrained least squares Levenberg optimizer for dense problems.

Parameters:

dampInit - Initial value of dampening parameter. Tune. Start at around 1e-3.

Returns:

UnconstrainedLeastSquares

leastSquaresTrustRegion

public static UnconstrainedLeastSquares leastSquaresTrustRegion(double regionSize,
                                                                RegionStepType type,
                                                                boolean robustSolver)

Creates a trust region based optimization algorithm for least squares problem.

Parameters:

regionSize - Maximum radius of the trust region.

type - The type of trust region

robustSolver - If true then a lower but more robust solver which can handle singularities, otherwise a much faster one if used.

Returns:

UnconstrainedLeastSquares

See Also:

TrustRegionLeastSquares