Package pulse.search.linear

Class WolfeOptimiser

java.lang.Object

pulse.util.UpwardsNavigable

pulse.util.Group

pulse.util.Accessible

pulse.util.PropertyHolder

pulse.search.linear.LinearOptimiser

pulse.search.linear.WolfeOptimiser

All Implemented Interfaces:

Serializable, Descriptive, Reflexive

public class WolfeOptimiser
extends LinearOptimiser

This is the implementation of the strong Wolfe conditions for performing inexact linear search. This type of linear search works best with the ApproximatedHessianSolver.

See Also:

BFGSOptimiser, Wikipedia page, Serialized Form

Field Summary

Fields

Modifier and Type	Field	Description
static double	C1	The constant used in the Armijo inequality, equal to 0.05.
static double	C2	The constant used in the strong Wolfe inequality for the modulus of the gradient projection, equal to 0.8.

Fields inherited from class pulse.search.linear.LinearOptimiser

searchResolution

Method Summary

Modifier and Type	Method	Description
static WolfeOptimiser	getInstance()	This class uses a singleton pattern, meaning there is only instance of this class.
double	linearStep(GeneralTask task)	This uses a combination of the Wolfe conditions for conducting an inexact line search with the domain partitioning using a random number generator.
String	toString()

Methods inherited from class pulse.search.linear.LinearOptimiser

domain, getLinearResolution, listedKeywords, set, setLinearResolution

Methods inherited from class pulse.util.PropertyHolder

addListener, areDetailsHidden, data, describe, firePropertyChanged, getDescriptor, getListeners, getPrefix, ignoreSiblings, initListeners, isListedNumericType, isListedParameter, listedTypes, numericData, parameterListChanged, removeListeners, setPrefix, updateProperties, updateProperty

Methods inherited from class pulse.util.Accessible

accessibleChildren, genericProperties, genericProperty, numericProperties, numericProperty, property, update, update

Methods inherited from class pulse.util.Group

access, children, contents, getSimpleName, subgroups

Methods inherited from class pulse.util.UpwardsNavigable

addHierarchyListener, getHierarchyListeners, getParent, identify, removeHierarchyListener, removeHierarchyListeners, setParent, specificAncestor, tellParent

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

C1

public static final double C1

The constant used in the Armijo inequality, equal to 0.05.

See Also:

Constant Field Values

C2

public static final double C2

The constant used in the strong Wolfe inequality for the modulus of the gradient projection, equal to 0.8.

See Also:

Constant Field Values

Method Detail

linearStep

public double linearStep(GeneralTask task)
                  throws SolverException

This uses a combination of the Wolfe conditions for conducting an inexact line search with the domain partitioning using a random number generator. The partitioning is done in such a way that: (a) whenever the Armijo inequality is not satisfied, the original domain [a; b] is reduced to $[$a_i; α], where α is the random number confined inside [a_i; b_i]; (b) when the Armijo inequality is satisfied and the second (strong) Wolfe condition for the modulus of the gradient projection is not satisfied, the α value is used to substitute the lower end point for the search domain: [α; b_i]. As this is done iteratively, the length of the associated Segment will decrease. The method will return a value if either the strong Wolfe conditions are strictly satisfied, or if the linear precision has been reached.

Specified by:

linearStep in class LinearOptimiser

Parameters:

task - the target function is the sum of squared residuals (SSR) for this task

Returns:

a double, representing the step magnitude that needs to be multiplied by the direction of the search determined previously using the PathSolver to arrive at the next set of parameters corresponding to a lower SSR value of this task

Throws:

SolverException

toString

public String toString()

Overrides:

toString in class LinearOptimiser

getInstance

public static WolfeOptimiser getInstance()

This class uses a singleton pattern, meaning there is only instance of this class.

Returns:

the single (static) instance of this class