Hi Sean,
Thanks a lot for you reply.
Please see my inline response.
On Thu, Jun 2, 2016 at 6:58 AM, Sean Luke <[log in to unmask]> wrote:
> On May 31, 2016, at 1:22 AM, Atm Golam Bari <[log in to unmask]> wrote:
>
> >>> MultiPopCoevolutionaryEvaluator (under ec/coevolve/) extends
> SteadyStateEvaluator and overrides evaluatePopulation(, , )
>
> 1. I presume you modified MultiPopCoevlutionaryEvaluator to do these
> things.
>
-Yes.
>
> 2. SteadyStateEvaluator doesn't use evaluatePopulation. It uses
> evaluateIndividual and works with SteadyStateEvolutionState, not
> SimpleEvolutionState.
>
-Yes
> >>> GroupedProblem inside MultiPopCoevolutionaryEvlauator is changed into
> SimpleProblem
>
> ??? The purpose of coevolution is to evaluate individuals in the context
> of one another. SimpleProblem would make no sense.
>
- My idea was like this : send one individual to the problem, play the
interaction game with other individuals from different population inside
problem class.
>
> >>> Inside performCoevolutionaryEvaluation (, , ), writing
> prob.evluate(....) is called two times each for one population to be
> workable with SimpleProblem.
>
> This isn't coevolution then. It's just two separate populations.
>
- I used the above idea to measure the context of one other inside the
problem class. I am not sure how much it make sense in terms of
coevolution.
>
> > What am I missing? Is it a right way to handle SSEA in a two pop
> coevoluton? Any suggestion or new approach is highly appreciated.
>
> I think you need to first think about what SSEA means in a coevolutionary
> context. Steady State is basically this:
>
>
> Loop forever:
> I <- breed new individual
> F <- evaluate individual
> G <- add individual into population, perhaps displacing an
> existing individual
>
> How would you modify this loop to do SSEA in the way you're hoping? What
> kind of coevolution are you trying to do?
>
- I need "Parallel 2-Population Competitive Coevolution"
- I don't want to modify this loop.
What I did in custom evaluator is as follows -
public class EvalParetoSSEA extends SteadyStateEvaluator {
---
public void evaluateIndividual(final EvolutionState state, Individual ind ,
int subpop) { //Overriding
...
performCoevolutionaryEvaluation( state, state.population,
(SimpleProblemForm)p_problem )
//See this is SimpleProblem, If I use GroupProblem I get a runtime error :
SSEA uses SimpleProblem
...
//This is what I did when want to send
from inside performCoevolutionaryEvluation
for(int i = 0; i < state.population.subpops[0].individuals.length; i++)
for(int k = 0; k < numShuffled; k++)
{
for(int ind = 0; ind < inds.length; ind++)
{
inds[ind] =
state.population.subpops[ind].individuals[ordering[k][ind][i]];
updates[ind] = true;
if (ind == 0)
prob.evaluate(state,inds[ind],subpops[ind], 0);
*//This line gets the
NullPointerException*
else
prob.evaluate(state,inds[ind],subpops[ind], 0);
}
//prob.evaluate(state,inds,updates, false, subpops, 0);
evaluations++;
}
}
}
}
So, the question that I want to ask now :
>>How shall I modify SimpleProblem to work with GroupProblem inside
performCoevolutionaryEvolution? Or do I really need it.
>>evaluteindividual of steadystateEvaluator sends one individual, if I
override evaluate individual inside custom evaluator I get null pointer
exception (Please see above). So, I can't use this version?
>> What I finally did is - use steadystateevaluator's evaluateindividual to
send individual from a single population one by one to the problem class,
catch that individual and the opponents from different population and
perform interaction game, set fitness etc. But I am not sure whether it is
any kind (say Parallel 2-pop competitive) of coevolution ?
Thanks a lot for your time.
> Sean
>
