Package ec.vector
Class FloatVectorSpecies
java.lang.Object
ec.Species
ec.vector.VectorSpecies
ec.vector.FloatVectorSpecies
- All Implemented Interfaces:
Prototype,Setup,Serializable,Cloneable
- Direct Known Subclasses:
AMALGAMSpecies,CMAESSpecies
FloatVectorSpecies is a subclass of VectorSpecies with special
constraints for floating-point vectors, namely FloatVectorIndividual and
DoubleVectorIndividual.
FloatVectorSpecies can specify a number of parameters globally, per-segment, and per-gene. See VectorSpecies for information on how to this works.
FloatVectorSpecies defines a minimum and maximum gene value. These values are used during initialization and, depending on whether mutation-bounded is true, also during various mutation algorithms to guarantee that the gene value will not exceed these minimum and maximum bounds.
FloatVectorSpecies provides support for five ways of mutating a gene.
- reset Replacing the gene's value with a value uniformly drawn from the gene's range (the default behavior).
- gaussPerturbing the gene's value with gaussian noise; if the gene-by-gene range is used, than the standard deviation is scaled to reflect each gene's range. If the gaussian mutation's standard deviation is too large for the range, than there's a large probability the mutated value will land outside range. We will try again a number of times (100) before giving up and using the previous mutation method.
- polynomial Perturbing the gene's value with noise chosen from a polynomial distribution, similar to the gaussian distribution. The polynomial distribution was popularized by Kalyanmoy Deb and is found in many of his publications (see http://www.iitk.ac.in/kangal/deb.shtml). The polynomial distribution has two options. First, there is the index. This variable defines the shape of the distribution and is in some sense the equivalent of the standard deviation in the gaussian distribution. The index is an integer. If it is zero, the polynomial distribution is simply the uniform distribution from [1,-1]. If it is 1, the polynomial distribution is basically a triangular distribution from [1,-1] peaking at 0. If it is 2, the polynomial distribution follows a squared function, again peaking at 0. Larger values result in even more peaking and narrowness. The default values used in nearly all of the NSGA-II and Deb work is 20. Second, there is whether or not the value is intended for bounded genes. The default polynomial distribution is used when we assume the gene can take on literally any value, even beyond the min and max values. For genes which are restricted to be between min and max, there is an alternative version of the polynomial distribution, used by Deb's team but not discussed much in the literature, desiged for that situation. We assume boundedness by default, and have found it to be somewhat better for NSGA-II and SPEA2 problems. For a description of this alternative version, see "A Niched-Penalty Approach for Constraint Handling in Genetic Algorithms" by Kalyanmoy Deb and Samir Agrawal. Deb's default implementation bounds the result to min or max; instead ECJ's implementation of the polynomial distribution retries until it finds a legal value. This will be just fine for ranges like [0,1], but for smaller ranges you may be waiting a long time.
- integer-reset Replacing the gene's value with a value uniformly drawn from the gene's range but restricted to only integers.
- integer-random-walk Replacing the gene's value by performing a random walk starting at the gene value. The random walk either adds 1 or subtracts 1 (chosen at random), then does a coin-flip to see whether to continue the random walk. When the coin-flip finally comes up false, the gene value is set to the current random walk position.
FloatVectorSpecies provides support for two ways of initializing a gene. The initialization procedure is determined by the choice of mutation procedure as described above. If the mutation is floating-point (reset, gauss, polynomial), then initialization will be done by resetting the gene to uniformly chosen floating-point value between the minimum and maximum legal gene values, inclusive. If the mutation is integer (integer-reset, integer-random-walk), then initialization will be done by performing the same kind of reset, but restricting values to integers only.
Parameters
| base.min-gene or | ||
| base.segment.segment-number.min-gene or | ||
| base.min-gene.gene-number 0.0 <= double <= 1.0 |
(probability that a gene will get mutated over default mutation) | (the minimum gene value) |
| base.max-gene or | ||
| base.segment.segment-number.max-gene or | ||
| base.max-gene.gene-number 0.0 <= double <= 1.0 |
(probability that a gene will get mutated over default mutation) | (the maximum gene value) |
| base.mutation-type or | ||
| base.segment.segment-number.mutation-type or | ||
| base.mutation-prob.gene-number reset, gauss, polynomial, integer-reset, or integer-random-walk (default=reset) |
(the mutation type) | |
| base.mutation-stdev or | ||
| base.segment.segment-number.mutation-stdev or | ||
| base.mutation-stdev.gene-number double ≥ 0 |
(the standard deviation or the gauss perturbation) | |
| base.distribution-index or | ||
| base.segment.segment-number.distribution-index or | ||
| base.distribution-index.gene-number int ≥ 0 |
(the mutation distribution index for the polynomial mutation distribution) | |
| base.alternative-polynomial-version or | ||
| base.segment.segment-number.alternative-polynomial-version or | ||
| base.alternative-polynomial-version.gene-number boolean (default=true) |
(whether to use the "bounded" variation of the polynomial mutation or the standard ("unbounded") version) | |
| base.random-walk-probability or | ||
| base.segment.segment-number.random-walk-probability or | ||
| base.random-walk-probability.gene-number 0.0 <= double <= 1.0 |
(the probability that a random walk will continue. Random walks go up or down by 1.0 until the coin flip comes up false.) | |
| base.mutation-bounded or | ||
| base.segment.segment-number.mutation-bounded or | ||
| base.mutation-bounded.gene-number boolean (default=true) |
(whether mutation is restricted to only being within the min/max gene values. Does not apply to SimulatedBinaryCrossover (which is always bounded)) | |
| base.out-of-bounds-retries int ≥ 0 (default=100) |
(number of times the gaussian mutation got the gene out of range before we give up and reset the gene's value; 0 means "never give up") |
- See Also:
-
Field Summary
FieldsModifier and TypeFieldDescriptionstatic final intstatic final intstatic final intstatic final intstatic final intstatic final intprotected double[]Standard deviation for Gaussian Mutation, per gene.protected double[]Max-gene value, per gene.protected double[]Min-gene value, per gene.protected int[]The distribution index for Polynomial Mutation, per gene.protected boolean[]Whether mutation is bounded to the min- and max-gene values, per gene.protected int[]Mutation type, per gene.intThe number of times Polynomial Mutation or Gaussian Mutation retry for valid numbers until they get one.static final Stringstatic final Stringstatic final Stringstatic final Stringstatic final Stringstatic final Stringstatic final Stringstatic final Stringstatic final Stringprotected boolean[]Whether the Polynomial Mutation method is the "alternative" method, per gene.protected double[]The continuation probability for Integer Random Walk Mutation, per gene.static final Stringstatic final Stringstatic final Stringstatic final Stringstatic final StringFields inherited from class ec.vector.VectorSpecies
C_ANY_POINT, C_GEOMETRIC, C_INTERMED_RECOMB, C_LINE_RECOMB, C_NONE, C_ONE_POINT, C_ONE_POINT_NO_NOP, C_SIMULATED_BINARY, C_TWO_POINT, C_TWO_POINT_NO_NOP, C_UNIFORM, chunksize, crossoverDistributionIndex, crossoverProbability, crossoverType, duplicateRetries, dynamicInitialSize, genomeIncreaseProbability, genomeResizeAlgorithm, genomeSize, lineDistance, maxInitialSize, minInitialSize, mutationProbability, P_CHUNKSIZE, P_CROSSOVER_DISTRIBUTION_INDEX, P_CROSSOVERPROB, P_CROSSOVERTYPE, P_DUPLICATE_RETRIES, P_GENOMESIZE, P_GEOMETRIC_PROBABILITY, P_LINEDISTANCE, P_MUTATIONPROB, P_NUM_SEGMENTS, P_SEGMENT, P_SEGMENT_END, P_SEGMENT_START, P_SEGMENT_TYPE, P_UNIFORM_MAX, P_UNIFORM_MIN, P_VECTORSPECIES, V_ANY_POINT, V_GEOMETRIC, V_INTERMED_RECOMB, V_LINE_RECOMB, V_ONE_POINT, V_ONE_POINT_NO_NOP, V_SIMULATED_BINARY, V_TWO_POINT, V_TWO_POINT_NO_NOP, V_UNIFORMFields inherited from class ec.Species
f_prototype, i_prototype, P_FITNESS, P_INDIVIDUAL, P_PIPE, pipe_prototype -
Constructor Summary
Constructors -
Method Summary
Modifier and TypeMethodDescriptiondoublegaussMutationStdev(int gene) booleaninNumericalTypeRange(double geneVal) protected voidloadParametersForGene(EvolutionState state, int index, Parameter base, Parameter def, String postfix) Called when VectorSpecies is setting up per-gene and per-segment parameters.doublemaxGene(int gene) doubleminGene(int gene) intmutationDistributionIndex(int gene) booleanmutationIsBounded(int gene) intmutationType(int gene) voidbooleanpolynomialIsAlternative(int gene) doublerandomWalkProbability(int gene) voidsetup(EvolutionState state, Parameter base) The default version of setup(...) loads requested pipelines and calls setup(...) on them and normalizes their probabilities.Methods inherited from class ec.vector.VectorSpecies
contains, contains, contains, contains, defaultBase, duplicateRetries, fill, fill, fill, fill, initializeGenomeSegmentsByEndIndices, initializeGenomeSegmentsByStartIndices, mutationProbability, newIndividual, setupGenomeMethods inherited from class ec.Species
buildMisc, clone, newIndividual, newIndividual, updateIndividual, updateSubpopulation
-
Field Details
-
P_MINGENE
- See Also:
-
P_MAXGENE
- See Also:
-
P_MUTATIONTYPE
- See Also:
-
P_STDEV
- See Also:
-
P_MUTATION_DISTRIBUTION_INDEX
- See Also:
-
P_POLYNOMIAL_ALTERNATIVE
- See Also:
-
V_RESET_MUTATION
- See Also:
-
V_GAUSS_MUTATION
- See Also:
-
V_POLYNOMIAL_MUTATION
- See Also:
-
V_INTEGER_RANDOM_WALK_MUTATION
- See Also:
-
V_INTEGER_RESET_MUTATION
- See Also:
-
P_RANDOM_WALK_PROBABILITY
- See Also:
-
P_OUTOFBOUNDS_RETRIES
- See Also:
-
P_MUTATION_BOUNDED
- See Also:
-
C_RESET_MUTATION
public static final int C_RESET_MUTATION- See Also:
-
C_GAUSS_MUTATION
public static final int C_GAUSS_MUTATION- See Also:
-
C_POLYNOMIAL_MUTATION
public static final int C_POLYNOMIAL_MUTATION- See Also:
-
C_INTEGER_RESET_MUTATION
public static final int C_INTEGER_RESET_MUTATION- See Also:
-
C_INTEGER_RANDOM_WALK_MUTATION
public static final int C_INTEGER_RANDOM_WALK_MUTATION- See Also:
-
minGene
protected double[] minGeneMin-gene value, per gene. This array is one longer than the standard genome length. The top element in the array represents the parameters for genes in genomes which have extended beyond the genome length. -
maxGene
protected double[] maxGeneMax-gene value, per gene. This array is one longer than the standard genome length. The top element in the array represents the parameters for genes in genomes which have extended beyond the genome length. -
mutationType
protected int[] mutationTypeMutation type, per gene. This array is one longer than the standard genome length. The top element in the array represents the parameters for genes in genomes which have extended beyond the genome length. -
gaussMutationStdev
protected double[] gaussMutationStdevStandard deviation for Gaussian Mutation, per gene. This array is one longer than the standard genome length. The top element in the array represents the parameters for genes in genomes which have extended beyond the genome length. -
mutationIsBounded
protected boolean[] mutationIsBoundedWhether mutation is bounded to the min- and max-gene values, per gene. This array is one longer than the standard genome length. The top element in the array represents the parameters for genes in genomes which have extended beyond the genome length. -
mutationDistributionIndex
protected int[] mutationDistributionIndexThe distribution index for Polynomial Mutation, per gene. This array is one longer than the standard genome length. The top element in the array represents the parameters for genes in genomes which have extended beyond the genome length. -
polynomialIsAlternative
protected boolean[] polynomialIsAlternativeWhether the Polynomial Mutation method is the "alternative" method, per gene. This array is one longer than the standard genome length. The top element in the array represents the parameters for genes in genomes which have extended beyond the genome length. -
randomWalkProbability
protected double[] randomWalkProbabilityThe continuation probability for Integer Random Walk Mutation, per gene. This array is one longer than the standard genome length. The top element in the array represents the parameters for genes in genomes which have extended beyond the genome length. -
outOfBoundsRetries
public int outOfBoundsRetriesThe number of times Polynomial Mutation or Gaussian Mutation retry for valid numbers until they get one. -
DEFAULT_OUT_OF_BOUNDS_RETRIES
public static final int DEFAULT_OUT_OF_BOUNDS_RETRIES- See Also:
-
-
Constructor Details
-
FloatVectorSpecies
public FloatVectorSpecies()
-
-
Method Details
-
outOfRangeRetryLimitReached
-
maxGene
public double maxGene(int gene) -
minGene
public double minGene(int gene) -
mutationType
public int mutationType(int gene) -
gaussMutationStdev
public double gaussMutationStdev(int gene) -
mutationIsBounded
public boolean mutationIsBounded(int gene) -
mutationDistributionIndex
public int mutationDistributionIndex(int gene) -
polynomialIsAlternative
public boolean polynomialIsAlternative(int gene) -
randomWalkProbability
public double randomWalkProbability(int gene) -
inNumericalTypeRange
public boolean inNumericalTypeRange(double geneVal) -
setup
Description copied from class:SpeciesThe default version of setup(...) loads requested pipelines and calls setup(...) on them and normalizes their probabilities. If your individual prototype might need to know special things about the species (like parameters stored in it), then when you override this setup method, you'll need to set those parameters BEFORE you call super.setup(...), because the setup(...) code in Species sets up the prototype. -
loadParametersForGene
protected void loadParametersForGene(EvolutionState state, int index, Parameter base, Parameter def, String postfix) Description copied from class:VectorSpeciesCalled when VectorSpecies is setting up per-gene and per-segment parameters. The index is the current gene whose parameter is getting set up. The Parameters in question are the bases for the gene. The postfix should be appended to the end of any parameter looked up (it often contains a number indicating the gene in question), such as state.parameters.exists(base.push(P_PARAM).push(postfix), def.push(P_PARAM).push(postfix)If you override this method, be sure to call super(...) at some point, ideally first.
- Overrides:
loadParametersForGenein classVectorSpecies
-