gov.sns.tools.optimization
Class RandomSearchAlgorithm.ShrinkSearcher

java.lang.Object
  |
  +--gov.sns.tools.optimization.RandomSearchAlgorithm.RandomSearcher
        |
        +--gov.sns.tools.optimization.RandomSearchAlgorithm.ShrinkSearcher

All Implemented Interfaces:

RandomSearchAlgorithm.Searcher, gov.sns.tools.optimization.ScoreMonitor

Enclosing class:

RandomSearchAlgorithm

protected class RandomSearchAlgorithm.ShrinkSearcher

extends RandomSearchAlgorithm.RandomSearcher

ShrinkSearcher searches for the next trial point by adjusting the search domain per variable depending on how much a variable has changed between the best solutions found so far. As the variables converge on a solution, the search space shrinks. The search space may also increase if a variable is not converging as we get closer to an optimal solution.

Field Summary

protected java.util.Map

variableWindows

Fields inherited from class gov.sns.tools.optimization.RandomSearchAlgorithm.RandomSearcher

changeProbabilityBase, randomGenerator, values

Constructor Summary

RandomSearchAlgorithm.ShrinkSearcher()
Constructor

Method Summary

protected void	buildWindows() Build the new bounds based upon the user specified domain and the search algorithm's search space per variable.
void	newTopSolution(TrialPoint oldPoint, TrialPoint newPoint) An event indicating that a new solution has been found which is better than the previous best solution according to the score given by the evaluator.
protected double	proposeValue(Variable variable) Propose a new value for the variable by selecting a random value from the variable's shrunken search space.
void	reset() reset for searching from scratch; forget history

Methods inherited from class gov.sns.tools.optimization.RandomSearchAlgorithm.RandomSearcher

newScore, nextPoint, nextPoint, nextTrialPoint

Methods inherited from class java.lang.Object

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

Field Detail

variableWindows

protected java.util.Map variableWindows

Constructor Detail

RandomSearchAlgorithm.ShrinkSearcher

public RandomSearchAlgorithm.ShrinkSearcher()

Constructor

Method Detail

reset

public void reset()

reset for searching from scratch; forget history

Specified by:

reset in interface RandomSearchAlgorithm.Searcher

Overrides:

reset in class RandomSearchAlgorithm.RandomSearcher

newTopSolution

public void newTopSolution(TrialPoint oldPoint,
                           TrialPoint newPoint)

An event indicating that a new solution has been found which is better than the previous best solution according to the score given by the evaluator.

Specified by:

newTopSolution in interface RandomSearchAlgorithm.Searcher

Overrides:

newTopSolution in class RandomSearchAlgorithm.RandomSearcher

Parameters:

oldPoint - The old best point.

newPoint - The new best point.

buildWindows

protected void buildWindows()

Build the new bounds based upon the user specified domain and the search algorithm's search space per variable. The search windows become the intersection of the user specified search domain and the algorithm's search space.

proposeValue

protected double proposeValue(Variable variable)

Propose a new value for the variable by selecting a random value from the variable's shrunken search space.

Overrides:

proposeValue in class RandomSearchAlgorithm.RandomSearcher

Parameters:

variable - the variable for which to propose a new value

Returns:

the new value to propose for the variable