Eric,
Thanks a lot. I found the missing dot.
I also can't reproduce the result now :)
NB: The random order was for the evaluator class. Basically, I was using
variant of CompetitivEvaluator
and found that randomizeOrder inside that CompetitiveEvalauator is the
culprit here.
Thanks again for your time.
-agb
On Wed, Feb 22, 2017 at 12:41 PM, Eric 'Siggy' Scott <[log in to unmask]>
wrote:
> Atm,
>
> I've been unable to reproduce this behavior. I've copied your custom
> produce() method into the OurVectorMutator class of Tutorial 2, and run it
> with very similar parameters to your own—but the children show up in the
> new population in the same order as their parents in the old population,
> which is what you're saying you expected (see attached screenshots).
>
> Does your custom mutator override any other methods?
>
> Siggy
>
>
> On Wed, Feb 22, 2017 at 9:50 AM, BARI, ATM GOLAM <[log in to unmask]>
> wrote:
>
>> Thanks a lot for your reply.
>>
>> What exactly are you doing in your VectorCrossoverPipeline.produce()
>> method?
>> - Nothing. I just put it as it is.(see the param file please)
>> I put chunk-size = genome-size in the param file which I understand does
>> not execute the loop inside defaultCrossover() responsible for respective
>> crosssover type operation. As a result, no cross-over.
>>
>> If you are still pulling individuals from the source, then your
>> population might be being affected by the TournamentSelection object.
>> - why it should be affected if I just pull the individuals from the
>> source (may be pipeline or selection method) and do nothing operation on
>> them but forward only. This is what I am exactly doing.
>>
>> If you want the operators to do truly *nothing* to the population, I
>> believe (after just glancing at ECJ's code) you can do true cloning by
>> having your produce() method call Breeder.reproduce() and setting the
>> 'produceChildrenFromSource' parameter to 'false'.
>> - Is not that an extra work... produce() inside
>> ec.vector.breed.VectroCrossooverPipeline() and
>> ec.vector.Breed.VectorMutationPipeline() gets the individuals from their
>> source and stick them into inds[q] location. So, if I want to do "nothing"
>> in that individuals , Ill forward them as they are. Please let me correct
>> if I am wrong. Please see the attached output. I get the individuals
>> untouched but their indexes are not same from old to new population.
>>
>> Please see the produce() method for mutator class below. I just point
>> old individual as parent and new individual as child whose ie. q value
>> index are same. But when this produce() hands over all the individuals to
>> the breeder, the individuals remain intact but their indexes does not match.
>>
>> public int produce(final int min,
>> final int max,
>> final int start,
>> final int subpopulation,
>> final Individual[] inds,
>> final EvolutionState state,
>> final int thread) {
>> int n = sources[0].produce(min,max,start,subpopulation,inds,state,
>> thread);
>>
>>
>> if (!state.random[thread].nextBoolean(likelihood))
>> return reproduce(n, start, subpopulation, inds, state, thread, false);
>>
>> if (!(sources[0] instanceof BreedingPipeline))
>> for(int q=start;q<n+start;q++)
>> inds[q] = (Individual)(inds[q].clone());
>>
>> if (!(inds[start] instanceof IntegerVectorIndividual))
>> state.output.fatal("OurMutatorPipeline didn't get an
>> IntegerVectorIndividual." +
>> "The offending individual is in subpopulation " + subpopulation + " and
>> it's:" + inds[start]);
>>
>> IntegerVectorSpecies species = (IntegerVectorSpecies)(inds[st
>> art].species);
>>
>>
>> IntegerVectorIndividual child, parent;
>> for(int q=start;q<n+start;q++) { //Iterates single time
>> //System.out.println("q " +q + " Individual Length " +inds.length);
>> child = (IntegerVectorIndividual)inds[q];
>> parent = (IntegerVectorIndividual)state.population.subpops[subpopulat
>> ion].individuals[q];
>> System.out.println("Before Mutation, "+ " child Individual :
>> "+child.genotypeToStringForHumans() + " parent Individual : "
>> +parent.genotypeToStringForHumans());
>>
>> for(int x=0;x<child.genome.length;x++) { //Does mutation
>> break;
>> //dummy
>> }
>> System.out.println("After Mutation, "+ " child Individual :
>> "+child.genotypeToStringForHumans() + " parent Individual : "
>> +parent.genotypeToStringForHumans());
>>
>> parent.evaluated=false;
>> }
>>
>> return n;
>> }
>> -agb
>>
>> On Wed, Feb 22, 2017 at 9:11 AM, Eric 'Siggy' Scott <[log in to unmask]>
>> wrote:
>>
>>> What exactly are you doing in your VectorCrossoverPipeline.produce()
>>> method? If you are still pulling individuals from the source, then your
>>> population might be being affected by the TournamentSelection object.
>>>
>>> If you want the operators to do truly *nothing* to the population, I
>>> believe (after just glancing at ECJ's code) you can do true cloning by
>>> having your produce() method call Breeder.reproduce() and setting the
>>> 'produceChildrenFromSource' parameter to 'false'.
>>>
>>> Siggy
>>>
>>> On Wed, Feb 22, 2017 at 8:44 AM, Atm Golam Bari <[log in to unmask]>
>>> wrote:
>>>
>>>> Hi,
>>>> I was just understanding how ECJ generational loop works. So, I
>>>> disabled the cross over using chunk-size = genome size, override
>>>> produce(....) of ec.vector.breed.VectorCrossoverPipeline and did not
>>>> do any mutation there. So, I expect the old and new individuals should be
>>>> same throughout the run. I am getting ok that part.
>>>>
>>>> >>But the index of the old individuals does not match with the index of
>>>> new individuals. Say individual [10, 20] has index 0 at old population but
>>>> it gets different index (say index 3) in new population.
>>>> >> I understand that newPop inside ec.simple.SimpleBreeder is the array
>>>> that holds new population and state.population.subpops[subpopulation].individuals[]
>>>> holds the old population at any time.
>>>>
>>>> Would you please let me know whether I am missing. Here is the part of
>>>> parameter file (attached file name : Capture.PNG) .
>>>> Thanks for your time,
>>>> agb
>>>>
>>>>
>>>
>>>
>>> --
>>>
>>> Ph.D student in Computer Science, George Mason University
>>> CFO and Web Director, Journal of Mason Graduate Research
>>> http://mason.gmu.edu/~escott8/
>>>
>>
>>
>>
>> --
>> -Bari
>>
>
>
>
> --
>
> Ph.D student in Computer Science, George Mason University
> CFO and Web Director, Journal of Mason Graduate Research
> http://mason.gmu.edu/~escott8/
>
--
-Bari
|
