ec.gp.koza.KozaFitness Maven / Gradle / Ivy
Show all versions of ecj Show documentation
/*
Copyright 2006 by Sean Luke
Licensed under the Academic Free License version 3.0
See the file "LICENSE" for more information
*/
package ec.gp.koza;
import ec.util.*;
import ec.*;
import java.io.*;
/*
* KozaFitness.java
*
* Created: Fri Oct 15 14:26:44 1999
* By: Sean Luke
*/
/**
* KozaFitness is a Fitness which stores an individual's fitness as described in
* Koza I. Well, almost. In KozaFitness, standardized fitness and raw fitness
* are considered the same (there are different methods for them, but they return
* the same thing). Standardized fitness ranges from 0.0 inclusive (the best)
* to infinity exclusive (the worst). Adjusted fitness converts this, using
* the formula adj_f = 1/(1+f), into a scale from 0.0 exclusive (worst) to 1.0
* inclusive (best). While it's the standardized fitness that is stored, it
* is the adjusted fitness that is printed out.
* This is all just convenience stuff anyway; selection methods
* generally don't use these fitness values but instead use the betterThan
* and equalTo methods.
*
Default Base
gp.koza.fitness
*
*
* @author Sean Luke
* @version 1.0
*/
public class KozaFitness extends Fitness
{
public static final String P_KOZAFITNESS = "fitness";
/** This ranges from 0 (best) to infinity (worst). I
define it here as equivalent to the standardized fitness. */
protected double standardizedFitness;
/** This auxillary measure is used in some problems for additional
information. It's a traditional feature of Koza-style GP, and so
although I think it's not very useful, I'll leave it in anyway. */
public int hits;
public Parameter defaultBase()
{
return GPKozaDefaults.base().push(P_KOZAFITNESS);
}
/**
Do not use this function. Use the identical setStandardizedFitness() instead.
The reason for the name change is that fitness() returns a differently-defined
value than setFitness() sets, ugh.
@deprecated
*/
public void setFitness(final EvolutionState state, final double _f)
{
setStandardizedFitness(state,_f);
}
/** Set the standardized fitness in the half-open interval [0.0,infinity)
which is defined (NOTE: DIFFERENT FROM fitness()!!!) as 0.0
being the IDEAL and infinity being worse than the worst possible.
This is the GP tradition. The fitness() function instead will output
the equivalent of Adjusted Fitness.
*/
public void setStandardizedFitness(final EvolutionState state, final double _f)
{
if (_f < 0.0 || _f >= Double.POSITIVE_INFINITY || Double.isNaN(_f))
{
state.output.warning("Bad fitness (may not be < 0, NaN, or infinity): " + _f + ", setting to 0.");
standardizedFitness = 0;
}
else standardizedFitness = _f;
}
/** Returns the adjusted fitness metric, which recasts the
fitness to the half-open interval (0,1], where 1 is ideal and
0 is worst. Same as adjustedFitness(). */
public double fitness()
{
return 1.0/(1.0 + standardizedFitness);
}
/** Returns the raw fitness metric.
@deprecated use standardizedFitness()
*/
public double rawFitness()
{
return standardizedFitness();
}
/** Returns the standardized fitness metric. */
public double standardizedFitness()
{
return standardizedFitness;
}
/** Returns the adjusted fitness metric, which recasts the fitness
to the half-open interval (0,1], where 1 is ideal and 0 is worst.
This metric is used when printing the fitness out. */
public double adjustedFitness()
{
return fitness();
}
public void setup(final EvolutionState state, final Parameter base) { }
public boolean isIdealFitness()
{
return standardizedFitness <= 0.0; // should always be == 0.0, <0.0 is illegal, but just in case...
}
public boolean equivalentTo(final Fitness _fitness)
{
// We're comparing standardized fitness because adjusted fitness can
// loose some precision in the division.
return ((KozaFitness)_fitness).standardizedFitness() == standardizedFitness;
}
public boolean betterThan(final Fitness _fitness)
{
// I am better than you if my standardized fitness is LOWER than you
// (that is, closer to zero, which is optimal)
// We're comparing standardized fitness because adjusted fitness can
// loose some precision in the division.
return ((KozaFitness)_fitness).standardizedFitness() > standardizedFitness;
}
public String fitnessToString()
{
return FITNESS_PREAMBLE + Code.encode(standardizedFitness) + Code.encode(hits);
}
public String fitnessToStringForHumans()
{
return FITNESS_PREAMBLE + "Standardized=" + standardizedFitness + " Adjusted=" + adjustedFitness() + " Hits=" + hits;
}
public void readFitness(final EvolutionState state,
final LineNumberReader reader)
throws IOException
{
DecodeReturn d = Code.checkPreamble(FITNESS_PREAMBLE, state, reader);
// extract fitness
Code.decode(d);
if (d.type!=DecodeReturn.T_DOUBLE)
state.output.fatal("Reading Line " + d.lineNumber + ": " +
"Bad Fitness.");
standardizedFitness = (double)d.d;
// extract hits
Code.decode(d);
if (d.type!=DecodeReturn.T_INT)
state.output.fatal("Reading Line " + d.lineNumber + ": " +
"Bad Fitness.");
hits = (int)d.l;
}
public void writeFitness(final EvolutionState state,
final DataOutput dataOutput) throws IOException
{
dataOutput.writeDouble(standardizedFitness);
dataOutput.writeInt(hits);
writeTrials(state, dataOutput);
}
public void readFitness(final EvolutionState state,
final DataInput dataInput) throws IOException
{
standardizedFitness = dataInput.readDouble();
hits = dataInput.readInt();
readTrials(state, dataInput);
}
public void setToMeanOf(EvolutionState state, Fitness[] fitnesses)
{
// this is not numerically stable. Perhaps we should have a numerically stable algorithm for sums
// we're presuming it's not a very large number of elements, so it's probably not a big deal,
// since this function is meant to be used mostly for gathering trials together.
double f = 0;
long h = 0;
for(int i = 0; i < fitnesses.length; i++)
{
KozaFitness fit = (KozaFitness)(fitnesses[i]);
f += fit.standardizedFitness;
h += fit.hits;
}
f /= fitnesses.length;
h /= fitnesses.length;
standardizedFitness = (double)f;
hits = (int)h;
}
}