|
»
»
»
keeping the evolve() population size constant
|
View:
New views
2 Messages
—
Rating Filter:
Alert me
|
 |
keeping the evolve() population size constant
by dave johnson
::
Rate this Message:
HI! I'm loving JGAP. It has been very helpful for some projects I've been working on.
I have a question concerning the ability to maintain a constant population size for the evolve() method. I am concerned because according to the javadoc, my own examination of the source code, and empirical tests, the population size does not stay constant when I call the evolve() method on my Genotype population. This is troubling because my BulkFitnessFunction evaluates the fitness of the chromosomes by running them through a tournament. Unfortunately, the tournament runs in O(n!) time (running time grows based on the factorial of n, where n is the poulation size). I am concerned because if I set my population size to 10, the evolve() method may end up calling my evaluate method of my BulkFitnessFunction with a population of 20 (or more) chromosomes. The difference between 10! = 3628800 and 20! = 2432902008176640000 is enormous and makes running my tournament impossible -- especially if I want to evolve() my population more than once! ;) Is there anything I can do (short of editing the source) to ensure that my population size will remain constant based on my configuration? I have already tried using setKeepPopulationSizeConstant (true)
on my configuration. I know I could also set my population size
smaller to limit the size of the population sent to my evaluate()
method, but that seems hackish. Is there anything else I can try? Maybe I am missing something? Below is some sample code and output to demonstrate my issue. Thanks in advance and keep up the great work! From testGA.java: public static void main(String[] args) throws InvalidConfigurationException { Configuration conf = new DefaultConfiguration(); Gene[] sampleGenes = new Gene[CHROMOSOME_SIZE]; for (int i = 0; i < CHROMOSOME_SIZE; i++) { sampleGenes[i] = new BooleanGene(conf); } Chromosome sampleChromosome = new Chromosome(conf, sampleGenes); conf.setSampleChromosome(sampleChromosome); conf.setPopulationSize(POPULATION_SIZE); conf.setKeepPopulationSizeConstant(true); conf.setBulkFitnessFunction(new MyBulkFitnessFunction()); Genotype population = Genotype.randomInitialGenotype(conf); for (int i = 0; i < 5; i++) { System.out.println("Genotype population size: " + Integer.toString(population.getPopulation().size())); population.evolve(); } } From MyBulkFitnessFunction.java: public void evaluate(Population pop) { Configuration a_config = pop.getConfiguration(); System.out.println(" Generation number: " + Integer.toString(a_config.getGenerationNr())); System.out.println(" Population size: " + Integer.toString(pop.size())); } Sample output: Genotype population size: 10 Generation number: 0 Population size: 19 Genotype population size: 10 Generation number: 1 Population size: 21 Genotype population size: 10 Generation number: 2 Population size: 22 Genotype population size: 10 Generation number: 3 Population size: 20 Genotype population size: 10 Generation number: 4 Population size: 18 -- dave ------------------------------------------------------------------------- Sponsored by: SourceForge.net Community Choice Awards: VOTE NOW! Studies have shown that voting for your favorite open source project, along with a healthy diet, reduces your potential for chronic lameness and boredom. Vote Now at http://www.sourceforge.net/community/cca08 _______________________________________________ jgap-users mailing list jgap-users@... https://lists.sourceforge.net/lists/listinfo/jgap-users |
|
|
Re: keeping the evolve() population size constant
by Klaus Meffert-5
::
Rate this Message:
Dave,
after investigating into the issue, I get the following
picture:
+ The population size stays constant at a certain point of
time in evolution (what you called "genotype population
size")
+ The population size varies regarding the input to the
bulk fitness function (what you called "population size")
+ The number of chromosomes in the population given to the
bulk fitness function, that are marked with operatedOn = 0, remains constant (at
least this should be the case and some few manual tests showed - although not
verified - that)
So, only the chromosomes that have a value > 0 as result
of call to method operatedOn (of interface IChromosome) are new ones regarding
the population. The other chromosomes are unchanged during the current evolution
cycle.
I wonder if this information could be used to reduce the
tournament size?
If not, then I offer the following
solution:
Modify class GABreeder, so that the code
block...
-- snip
BulkFitnessFunction
bulkFunction = a_conf.getBulkFitnessFunction();
if (bulkFunction != null) { /**@todo utilize jobs: bulk fitness function is not so important for a * prototype! */ bulkFunction.evaluate(pop); } -- snap ..is
moved down after the next statement ("pop = applyNaturalSelectors(a_conf, pop,
false);").
With
your test program, the output then shows a constant population size of
10.
Should
I add an option for executing the bulk function at the new position (with
default set in a way that executes the bulk function as before)? Is it dangerous
to execute the bulk function always in the new position?
Best
------------------------------------------------------------------------- Sponsored by: SourceForge.net Community Choice Awards: VOTE NOW! Studies have shown that voting for your favorite open source project, along with a healthy diet, reduces your potential for chronic lameness and boredom. Vote Now at http://www.sourceforge.net/community/cca08 _______________________________________________ jgap-users mailing list jgap-users@... https://lists.sourceforge.net/lists/listinfo/jgap-users |
| Free Forum Powered by Nabble | Forum Help |