public class MultiPopCoevolutionaryEvaluator extends Evaluator
MultiPopCoevolutionaryEvaluator is an Evaluator which performs competitive or cooperative multi-population coevolution. Competitive coevolution is where individuals' fitness is determined by testing them against individuals from other subpopulation. Cooperative coevolution is where individuals form teams together with members of other subpopulations, and the individuals' fitness is computed based on the performance of such teams. This evaluator assumes that the problem can only evaluate groups of individuals containing one individual from each subpopulation. Individuals are evaluated regardless of whether or not they've been evaluated in the past.
Your Problem is responsible for updating up the fitness appropriately with values usually obtained from teaming up the individual with different partners from the other subpopulations. MultiPopCoevolutionaryEvaluator expects to use Problems which adhere to the GroupedProblemForm interface, which defines a new evaluate(...) function, plus a preprocess(...) and postprocess(...) function.
This coevolutionary evaluator is single-threaded -- maybe we'll hack in multithreading later. It allows any number of subpopulations (implicitly, any number of individuals being evaluated together). The order of individuals in the subpopulation may be changed during the evaluation process.
Ordinarily MultiPopCoevolutionaryEvaluator does "parallel" coevolution: all subpopulations are evaluated simultaneously, then bred simultaneously. But if you set the "sequential" parameter in the class ec.simple.SimpleBreeder, then MultiPopCoevolutionary behaves in a sequential fashion common in the "classic" version of cooperative coevolution: only one subpopulation is evaluated and bred per generation. The subpopulation index to breed is determined by taking the generation number, modulo the total number of subpopulations.
Parameters
breed.sequential boolean (default = false) |
(should we evaluate and breed a single subpopulation each generation? Note that this is a SimpleBreeder parameter. ) |
base.subpop.num-current int >= 0 |
(the number of random individuals from any given subpopulation fropm the current population to be selected as collaborators) |
base.subpop.num-elites int >= 0 |
(the number of elite individuals from any given subpopulation from the previous population to be selected as collaborators. For generation 0, random individuals from the current population will be used. ) |
base.subpop.num-prev int >= 0 |
(the number of random individuals from any given subpopulation from the previous population to be selected as collaborators. For generation 0, random individuals from the current population will be used) |
base.subpop.X.select-prev instance of ec.SelectionMethod |
(the SelectionMethod used to select partners from the individuals in subpopulation X at the previous generation) |
base.subpop.X.select-current instance of ec.SelectionMethod |
(the SelectionMethod used to select partners from the individuals in subpopulation X at the current generation. WARNING. This SelectionMethod must not select based on fitness, since fitness hasn't been set yet. RandomSelection is a good choice. ) |
base.shuffling boolean (default = false) |
(instead of selecting individuals from ) |
Modifier and Type | Field and Description |
---|---|
protected int |
numCurrent |
protected int |
numElite |
protected int |
numPrev |
protected int |
numShuffled |
static java.lang.String |
P_NUM_ELITE |
static java.lang.String |
P_NUM_IND |
static java.lang.String |
P_NUM_RAND_IND |
static java.lang.String |
P_NUM_SHUFFLED |
static java.lang.String |
P_SELECTION_METHOD_CURRENT |
static java.lang.String |
P_SELECTION_METHOD_PREV |
static java.lang.String |
P_SUBPOP |
masterproblem, P_IAMSLAVE, P_MASTERPROBLEM, p_problem, P_PROBLEM, runComplete
Constructor and Description |
---|
MultiPopCoevolutionaryEvaluator() |
Modifier and Type | Method and Description |
---|---|
protected void |
afterCoevolutionaryEvaluation(EvolutionState state,
Population population,
GroupedProblemForm prob) |
protected void |
beforeCoevolutionaryEvaluation(EvolutionState state,
Population population,
GroupedProblemForm prob) |
void |
evaluatePopulation(EvolutionState state)
Evaluates the fitness of an entire population.
|
void |
performCoevolutionaryEvaluation(EvolutionState state,
Population population,
GroupedProblemForm prob) |
protected Individual |
produceCurrent(int subpopulation,
EvolutionState state,
int thread)
Selects one individual from the given subpopulation.
|
protected Individual |
producePrevious(int subpopulation,
EvolutionState state,
int thread)
Selects one individual from the previous subpopulation.
|
java.lang.String |
runComplete(EvolutionState state)
Returns non-NULL if the Evaluator believes that the run is
finished: perhaps an ideal individual has been found or some
other run result has shortcircuited the run so that it should
end prematurely right now.
|
void |
setup(EvolutionState state,
Parameter base)
Sets up the object by reading it from the parameters stored
in state, built off of the parameter base base.
|
boolean |
shouldEvaluateSubpop(EvolutionState state,
int subpop,
int threadnum)
Returns true if the subpopulation should be evaluated.
|
protected void |
shuffle(EvolutionState state,
int[] a) |
closeContacts, initializeContacts, reinitializeContacts, setRunComplete
public static final java.lang.String P_SUBPOP
public static final java.lang.String P_NUM_RAND_IND
protected int numCurrent
public static final java.lang.String P_NUM_SHUFFLED
protected int numShuffled
public static final java.lang.String P_NUM_ELITE
protected int numElite
public static final java.lang.String P_NUM_IND
protected int numPrev
public static final java.lang.String P_SELECTION_METHOD_PREV
public static final java.lang.String P_SELECTION_METHOD_CURRENT
public void setup(EvolutionState state, Parameter base)
Setup
public java.lang.String runComplete(EvolutionState state)
Evaluator
runComplete
in class Evaluator
public boolean shouldEvaluateSubpop(EvolutionState state, int subpop, int threadnum)
public void evaluatePopulation(EvolutionState state)
Evaluator
evaluatePopulation
in class Evaluator
protected void beforeCoevolutionaryEvaluation(EvolutionState state, Population population, GroupedProblemForm prob)
protected void shuffle(EvolutionState state, int[] a)
public void performCoevolutionaryEvaluation(EvolutionState state, Population population, GroupedProblemForm prob)
protected Individual producePrevious(int subpopulation, EvolutionState state, int thread)
protected Individual produceCurrent(int subpopulation, EvolutionState state, int thread)
protected void afterCoevolutionaryEvaluation(EvolutionState state, Population population, GroupedProblemForm prob)