Learn R
rgp (version 0.4-1)

makeEmptyRestartCondition: Evolution restart conditions

Description

Evolution restart conditions are predicates (functions that return a single logical value) of the signature function(population, fitnessFunction, stepNumber, evaluationNumber, bestFitness, timeElapsed). They are used to decide when to restart a GP evolution run that might be stuck in a local optimum. Evolution restart conditions are objects of the same type and class as evolution stop conditions. They may be freely substituted for each other.

Usage

makeEmptyRestartCondition()


makeStepLimitRestartCondition(stepLimit = 10)


makeFitnessStagnationRestartCondition(fitnessHistorySize = 100, testFrequency = 10, fitnessStandardDeviationLimit = 1e-06)


makeFitnessDistributionRestartCondition(testFrequency = 100, fitnessStandardDeviationLimit = 1e-06)

Arguments

stepLimit
The step limit for makeStepLimitRestartCondition.
fitnessHistorySize
The number of past best fitness values to look at when calculating the best fitness standard deviation for makeFitnessStagnationRestartCondition.
testFrequency
The frequency to test for the restart condition, in evolution steps. This parameter is mainly used with restart condititions that are expensive to calculate.
fitnessStandardDeviationLimit
The best fitness standard deviation limit for makeFitnessStagnationRestartCondition.

Details

makeEmptyRestartCondition creates a restart condition that is never fulfilled, i.e. restarts will never occur. makeStepLimitRestartCondition creates a restart condition that holds if the number if evolution steps is an integer multiple of a given step limit. restarts will never occur. makeFitnessStagnationRestartCondition creates a restart strategy that holds if the standard deviation of a last fitnessHistorySize best fitness values falls below a given fitnessStandardDeviationLimit. makeFitnessDistributionRestartCondition creates a restart strategy that holds if the standard deviation of the fitness values of the individuals in the current population falls below a given fitnessStandardDeviationLimit.