Package ec.de

Class DEBreeder

java.lang.Object
ec.Breeder
ec.de.DEBreeder
All Implemented Interfaces:
Setup, Singleton, Serializable
Direct Known Subclasses:
Best1BinDEBreeder, Rand1EitherOrDEBreeder
public class DEBreeder extends Breeder
DEBreeder provides a straightforward Differential Evolution (DE) breeder for the ECJ system. The code is derived from the "classic" DE algorithm, known as DE/rand/1/bin, found on page 140 of "Differential Evolution: A Practical Approach to Global Optimization" by Kenneth Price, Rainer Storn, and Jouni Lampinen.

DEBreeder requires that all individuals be DoubleVectorIndividuals.

In short, the algorithm is as follows. For each individual in the population, we produce a child by selecting three (different) individuals, none the original individual, called r0, r1, and r2. We then create an individal c, defined as c = r0 + F * (r1 - r2). Last, we cross over c with the original individual and produce a single child, using uniform crossover with gene-independent crossover probability "Cr".

This class should be used in conjunction with DEEvaluator, which allows the children to enter the population only if they're superior to their parents (the original individuals). If so, they replace their parents.

Parameters

base.f
0.0 <= double <= 1.0 		The "F" mutation scaling factor
base.cr
0.0 <= double <= 1.0 		The "Cr" probability of crossing over genes
See Also:

  • Field Details

    • CR_UNSPECIFIED

      public static final double CR_UNSPECIFIED
      See Also:

    • F

      public double F
      Scaling factor for mutation

    • Cr

      public double Cr
      Probability of crossover per gene

    • retries

      public int retries

    • P_F

      public static final String P_F
      See Also:

    • P_Cr

      public static final String P_Cr
      See Also:

    • P_OUT_OF_BOUNDS_RETRIES

      public static final String P_OUT_OF_BOUNDS_RETRIES
      See Also:

    • previousPopulation

      public Population previousPopulation
      the previous population is stored in order to have parents compete directly with their children

    • bestSoFarIndex

      public int[] bestSoFarIndex
      the best individuals in each population (required by some DE breeders). It's not required by DEBreeder's algorithm

  • Constructor Details

    • DEBreeder

      public DEBreeder()

  • Method Details

    • setup

      public void setup(EvolutionState state, Parameter base)
      Description copied from interface: Setup
      Sets up the object by reading it from the parameters stored in state, built off of the parameter base base. If an ancestor implements this method, be sure to call super.setup(state,base); before you do anything else.

    • prepareDEBreeder

      public void prepareDEBreeder(EvolutionState state)

    • breedPopulation

      public Population breedPopulation(EvolutionState state)
      Description copied from class: Breeder
      Breeds state.population, returning a new population. In general, state.population should not be modified.
      Specified by:
      breedPopulation in class Breeder

    • valid

      public boolean valid(DoubleVectorIndividual ind)
      Tests the Individual to see if its values are in range.

    • createIndividual

      public DoubleVectorIndividual createIndividual(EvolutionState state, int subpop, int index, int thread)

    • crossover

      public DoubleVectorIndividual crossover(EvolutionState state, DoubleVectorIndividual target, DoubleVectorIndividual child, int thread)
      Crosses over child with target, storing the result in child and returning it. The default procedure copies each value from the target, with independent probability CROSSOVER, into the child. The crossover guarantees that at least one child value, chosen at random, will not be overwritten. Override this method to perform some other kind of crossover.