halfdan/eva2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Added a reevaluation mechanism to DE which updates the fitness of old individuals instead of discarding them. This allows the optimization of noisy problems with DE

Browse Source
This commit is contained in:
Michael de Paly 2011-02-18 16:31:36 +00:00
parent 489e0a6f3f
commit bb7d87bafa
1 changed files with 53 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
src/eva2/server/go/strategies/DifferentialEvolution.java
View File

@ -41,6 +41,7 @@ public class DifferentialEvolution implements InterfaceOptimizer, java.io.Serial
private double m_Lambda = 0.6;
private double m_Mt = 0.05;
private int maximumAge = -1;
private boolean reEvaluate = false;
// to log the parents of a newly created indy.
private boolean doLogParents = false; // deactivate for better performance
private transient Vector<AbstractEAIndividual> parents = null;
@ -476,6 +477,19 @@ public class DifferentialEvolution implements InterfaceOptimizer, java.io.Serial
children.setGenerationTo(m_Population.getGeneration());
m_Problem.evaluate(children);
/**
* MdP: added a reevalutation mechanism for dynamically changing problems
*/
if(isReEvaluate()){
for(int i=0;i<this.m_Population.size();i++){
if (((AbstractEAIndividual)m_Population.get(i)).getAge() >= maximumAge) {
this.m_Problem.evaluate(((AbstractEAIndividual)m_Population.get(i)));
((AbstractEAIndividual)m_Population.get(i)).SetAge(0);
m_Population.incrFunctionCalls();
}
}
}
int nextDoomed = getNextDoomed(m_Population, 0);
for (int i = 0; i < this.m_Population.size(); i++) {
@ -511,6 +525,23 @@ public class DifferentialEvolution implements InterfaceOptimizer, java.io.Serial
// required for dynamic problems especially
m_Problem.evaluatePopulationStart(m_Population);
/**
* MdP: added a reevalutation mechanism for dynamically changing problems
*/
if(isReEvaluate()){
nextDoomed=-1;
for(int i=0;i<this.m_Population.size();i++){
if (((AbstractEAIndividual)m_Population.get(i)).getAge() >= maximumAge) {
this.m_Problem.evaluate(((AbstractEAIndividual)m_Population.get(i)));
((AbstractEAIndividual)m_Population.get(i)).SetAge(0);
m_Population.incrFunctionCalls();
}
}
}
for (int i = 0; i < this.m_Population.size(); i++) {
if (cyclePop) index=i;
else index=RNG.randomInt(0, this.m_Population.size()-1);
@ -654,7 +685,7 @@ public class DifferentialEvolution implements InterfaceOptimizer, java.io.Serial
return result;
}
/** This method allows you to set an identifier for the algorithm
* @param name The indenifier
* @param name The identifier
*/
public void SetIdentifier(String name) {
this.m_Identifier = name;
@ -685,8 +716,8 @@ public class DifferentialEvolution implements InterfaceOptimizer, java.io.Serial
return "DE";
}
/** Assuming that all optimizer will store thier data in a population
* we will allow acess to this population to query to current state
/** Assuming that all optimizer will store their data in a population
* we will allow access to this population to query to current state
* of the optimizer.
* @return The population of current solutions to a given problem.
*/
@ -873,4 +904,23 @@ public class DifferentialEvolution implements InterfaceOptimizer, java.io.Serial
public String cyclePopTipText() {
return "if true, individuals are used as parents in a cyclic sequence - otherwise randomly ";
}
/**
* @return the maximumAge
**/
public boolean isReEvaluate() {
return reEvaluate;
}
/**
* @param maximumAge the maximumAge to set
**/
public void setReEvaluate(boolean reEvaluate) {
this.reEvaluate = reEvaluate;
}
public String reEvaluateTipText() {
return "Reeavulates individuals which are older than maximum age instead of discarding them";
}
}