/*
Copyright 20010 by Sean Luke and George Mason University
Licensed under the Academic Free License version 3.0
See the file "LICENSE" for more information
*/
package ec.gp.ge;
import ec.*;
import ec.gp.*;
import ec.gp.koza.*;
import ec.simple.*;
import ec.coevolve.*;
import ec.util.*;
/*
* GEProblem.java
*
* Created: Sat Oct 16 23:21:01 EDT 2010
* By: Joseph Zelibor III, Eric Kangas, and Sean Luke
*/
/**
GEProblem is a special replacement for Problem which performs GE mapping. You do not subclass
from GEProblem. Rather, create a GPProblem subclass and set it to be the 'problem' parameter of the GEProblem.
The GEProblem will convert the GEIndividual into a GPIndividual, then pass this GPIndividual to the GPProblem
to be evaluated.
The procedure is as follows. Let's say your GPProblem is the Artificial Ant problem. Instead of saying...
eval.problem = ec.app.ant.Ant
eval.problem = ec.app.ant.Ant
eval.problem.data = ec.app.ant.AntData
eval.problem.moves = 400
eval.problem.file = santafe.trl
... you instead make your problem a GEProblem like this:
eval.problem = ec.gp.ge.GEProblem
... and then you hang the Ant problem, and all its subsidiary data, as the 'problem' parameter from the GEProblem like so:
eval.problem.problem = ec.app.ant.Ant
eval.problem.problem.data = ec.app.ant.AntData
eval.problem.problem.moves = 400
eval.problem.problem.file = santafe.trl
Everything else should be handled for you. GEProblem is also compatible with the MasterProblem procedure
for distributed evaluation, and is also both a SimpleProblemForm and a GroupedProblemForm. We've got you covered.
Parameters
base.problem
classname, inherits from GPProblem
(The GPProblem which actually performs the evaluation of the mapped GPIndividual)
*/
public class GEProblem extends Problem implements SimpleProblemForm, GroupedProblemForm
{
private static final long serialVersionUID = 1;
public final static String P_PROBLEM = "problem";
public GPProblem problem;
public void setup(EvolutionState state, Parameter base)
{
problem = (GPProblem)state.parameters.getInstanceForParameter(base.push(P_PROBLEM), null, GPProblem.class);
problem.setup(state, base.push(P_PROBLEM));
}
public Object clone()
{
GEProblem other = (GEProblem)(super.clone());
other.problem = (GPProblem)(problem.clone());
return other;
}
public void prepareToEvaluate(final EvolutionState state, final int threadnum)
{
problem.prepareToEvaluate(state, threadnum);
}
public void finishEvaluating(final EvolutionState state, final int threadnum)
{
problem.finishEvaluating(state, threadnum);
}
public void initializeContacts( EvolutionState state )
{
problem.initializeContacts(state);
}
public void reinitializeContacts( EvolutionState state )
{
problem.reinitializeContacts(state);
}
public void closeContacts(EvolutionState state, int result)
{
problem.closeContacts(state, result);
}
public boolean canEvaluate()
{
return problem.canEvaluate();
}
public void preprocessPopulation(final EvolutionState state, Population pop, final boolean[] prepareForFitnessAssessment, boolean countVictoriesOnly)
{
if (!(problem instanceof GroupedProblemForm))
state.output.fatal("GEProblem's underlying Problem is not a GroupedProblemForm");
((GroupedProblemForm)problem).preprocessPopulation(state, pop, prepareForFitnessAssessment, countVictoriesOnly);
}
public void postprocessPopulation(final EvolutionState state, Population pop, boolean[] assessFitness, final boolean countVictoriesOnly)
{
((GroupedProblemForm)problem).preprocessPopulation(state, pop, assessFitness, countVictoriesOnly);
}
/** Default version assumes that every individual is a GEIndividual.
The underlying problem.evaluate() must be prepared for the possibility that some
GPIndividuals handed it are in fact null, meaning that they couldn't be extracted
from the GEIndividual string. You should assign them bad fitness in some appropriate way.
*/
public void evaluate(final EvolutionState state,
final Individual[] ind, // the individuals to evaluate together
final boolean[] updateFitness, // should this individuals' fitness be updated?
final boolean countVictoriesOnly, // don't bother updating Fitness with socres, just victories
final int[] subpops,
final int threadnum)
{
// the default version assumes that every subpopulation is a GE Individual
Individual[] gpi = new Individual[ind.length];
for(int i = 0; i < gpi.length; i++)
{
if (ind[i] instanceof GEIndividual)
{
GEIndividual indiv = (GEIndividual) ind[i];
GESpecies species = (GESpecies) (ind[i].species);
// warning: gpi[i] may be null
gpi[i] = species.map(state, indiv, threadnum, null);
}
else if (ind[i] instanceof GPIndividual)
{
state.output.warnOnce("GPIndividual provided to GEProblem. Hope that's correct.");
gpi[i] = ind[i];
}
else
{
state.output.fatal("Individual " + i + " passed to Grouped evaluate(...) was neither a GP nor GE Individual: " + ind[i]);
}
}
((GroupedProblemForm)problem).evaluate(state, gpi, updateFitness, countVictoriesOnly, subpops, threadnum);
for(int i = 0; i < gpi.length; i++)
{
// Now we need to move the evaluated flag from the GPIndividual
// to the GEIndividual, and also for good measure, let's copy over
// the GPIndividual's fitness because even though the mapping function
// set the two Individuals to share the same fitness, it's possible
// that the evaluation function may have replaced the fitness.
ind[i].fitness = gpi[i].fitness; // if it's a GPIndividual anyway it'll just copy onto itself
ind[i].evaluated = gpi[i].evaluated; // if it's a GPIndividual anyway it'll just copy onto itself
}
}
public void evaluate(final EvolutionState state,
final Individual ind,
final int subpopulation,
final int threadnum)
{
// this shouldn't ever happen because GEProblem's Problems are ALWAYS
// SimpleProblemForm, but we include it here to be future-proof
if (!(problem instanceof SimpleProblemForm))
state.output.fatal("GEProblem's underlying Problem is not a SimpleProblemForm");
if (ind instanceof GEIndividual)
{
GEIndividual indiv = (GEIndividual) ind;
GESpecies species = (GESpecies) (ind.species);
GPIndividual gpi = species.map(state, indiv, threadnum, null);
if (gpi == null)
{
KozaFitness fitness = (KozaFitness) (ind.fitness);
fitness.setStandardizedFitness(state, Double.MAX_VALUE);
ind.fitness = fitness ;
ind.evaluated = true ;
}
else
{
((SimpleProblemForm)problem).evaluate(state, gpi, subpopulation, threadnum);
// Now we need to move the evaluated flag from the GPIndividual
// to the GEIndividual, and also for good measure, let's copy over
// the GPIndividual's fitness because even though the mapping function
// set the two Individuals to share the same fitness, it's possible
// that the evaluation function may have replaced the fitness.
ind.fitness = gpi.fitness;
ind.evaluated = gpi.evaluated;
}
}
else if (ind instanceof GPIndividual)
{
state.output.warnOnce("GPIndividual provided to GEProblem. Hope that's correct.");
((SimpleProblemForm)problem).evaluate(state, ind, subpopulation, threadnum); // just evaluate directly
}
else
{
state.output.fatal("Individual passed to evaluate(...) was neither a GP nor GE Individual: " + ind);
}
}
public void describe(final EvolutionState state,
final Individual ind,
final int subpopulation,
final int threadnum,
final int log)
{
if (ind instanceof GEIndividual)
{
GEIndividual indiv = (GEIndividual) ind;
GESpecies species = (GESpecies) (ind.species);
GPIndividual gpi = species.map(state, indiv, threadnum, null);
if (gpi != null)
{
problem.describe(state, gpi, subpopulation, threadnum, log);
// though this is probably not necessary for describe(...),
// for good measure we're doing the same rigamarole that we
// did for evaluate(...) above.
ind.fitness = gpi.fitness;
ind.evaluated = gpi.evaluated;
}
}
else if (ind instanceof GPIndividual)
{
state.output.warnOnce("GPIndividual provided to GEProblem. Hope that's correct.");
((SimpleProblemForm)problem).describe(state, ind, subpopulation, threadnum, log); // just describe directly
}
else
{
state.output.fatal("Individual passed to describe(...) was neither a GP nor GE Individual: " + ind);
}
}
}
Ads help maintain this website.