Genetic Search Algorithms Matt Herbster. Why Another Search? Designed in the 1950s, heavily...

12
Genetic Search Genetic Search Algorithms Algorithms Matt Herbster Matt Herbster

description

Definitions  Chromosome  Gene  Allele  Locus  Genotype  Phenotype  String  Feature, character  Feature value  String position  Structure  Parameter set

Transcript of Genetic Search Algorithms Matt Herbster. Why Another Search? Designed in the 1950s, heavily...

Page 1: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

Genetic Search Genetic Search AlgorithmsAlgorithmsMatt HerbsterMatt Herbster

Page 2: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

Why Another Search?Why Another Search? Designed in the 1950s, heavily Designed in the 1950s, heavily

implemented under John Holland (1970s)implemented under John Holland (1970s) Genetic search is intended to simulate Genetic search is intended to simulate

natural systemsnatural systems Works best on continuous and discrete Works best on continuous and discrete

combinatorial problemscombinatorial problems It will only take eight minutes of your timeIt will only take eight minutes of your time

Page 3: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

DefinitionsDefinitions ChromosomeChromosome GeneGene AlleleAllele LocusLocus GenotypeGenotype PhenotypePhenotype

StringString Feature, characterFeature, character Feature valueFeature value String positionString position StructureStructure Parameter setParameter set

Page 4: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

CharacteristicsCharacteristics ReproductionReproduction CrossoverCrossover MutationMutation

Rarely usedRarely used

Page 5: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

Mutation OperationsMutation Operations

Generative

Swap Node

Swap Sequence

Destructive

Page 6: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

Crossover OperationsCrossover Operations

Single pointOrder based

Page 7: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

Other RepresentationsOther Representations

Array, matrixArray, matrix TreeTree String of bitsString of bits Any other data structureAny other data structure

Page 8: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

ImplementationImplementation

1.1. Start with an initial gene poolStart with an initial gene pool2.2. Generate successors (either randomly Generate successors (either randomly

or deterministically) to create the first or deterministically) to create the first generation poolgeneration pool

3.3. Each node is evaluated by a fitness Each node is evaluated by a fitness function and sorted accordinglyfunction and sorted accordingly

4.4. Create new generations with the better Create new generations with the better most likely to reproducemost likely to reproduce

Page 9: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

What is the meaning of What is the meaning of the word better?the word better?

As determined by fitness function As determined by fitness function (essentially a heuristic)(essentially a heuristic)

Nodes with desired genes are Nodes with desired genes are predeterminedpredetermined

Can often approach local maxima rather Can often approach local maxima rather than the global optimal solutionthan the global optimal solution

Assisted by random-restart hill climbingAssisted by random-restart hill climbing

Page 10: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

VariationsVariations Genetic algorithms produce optimal Genetic algorithms produce optimal

results for many problems, eventually …results for many problems, eventually … Speciation – Two nodes will reproduce Speciation – Two nodes will reproduce

only if closely relatedonly if closely related Technique helps improve speedTechnique helps improve speed Parallel populations – simulates physical Parallel populations – simulates physical

separation with possible migrationseparation with possible migration

Page 11: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

ApplicationsApplications

Traveling salesman problemTraveling salesman problem Drilling of printed circuit boardsDrilling of printed circuit boards Planning bus routesPlanning bus routes

SchedulingScheduling Computer games – represent an Computer games – represent an

evolution of players’ strategiesevolution of players’ strategies Stock market trading – data fitting, trend Stock market trading – data fitting, trend

spotting, budgetingspotting, budgeting

Page 12: Genetic Search Algorithms Matt Herbster. Why Another Search?  Designed in the 1950s, heavily implemented under John Holland (1970s)  Genetic search.

QuestionsQuestions