A Generic Parallel Genetic Algorithm

1
A Generic Parallel Genetic Algorithm
By Roderick Murphy under the supervision of Mr
Dermot Frost
2
What Are Genetic Algorithms?

• Search or optimisation procedures based on the
  mechanisms of natural selection and natural
  genetics.
• i.e. the thoeries of this man

3
What Are Genetic Algorithms?

• They are weak optimisation techniques They
  dont use domain specific knowledge in their
  search procedure.
• They generally involve evolving a population of
  candidate solutions to a given problem.
• Evolution is carried out using operations
  inspired by natural genetic variation and natural
  selection.

4
Search Spaces and Fitness Landscapes
5
A Typical GA

• Random guesses of the solution to the problem
  An initial population.
• A means of calculating how good a guess solution
  is A fitness function.
• A method of mixing good solutions to produce
  better ones Crossover.
• An operator to introduce diversity within the
  population Mutation.

6
GA Terminology

• Chromosome / Genome String of characters
• Gene Characters used (eg binary)
• Allele 1 or 0 (for binary)
• Locus Position of gene in string
• Phenotype / Organism Candidate solution
• Generation Iteration

7
GA Operators

• There are 3 main operators for a serial GA

8
Selection

• The method by which
• population members
• (candidate solutions) are choosen.
• The chosen individuals will be combined with each
  other to form offspring.

9
Selection methods

• Common selection methods used in GAs are
• Fitness Proportionate Selection
• Rank Selection
• Tournament Selection

10
Fitness proportionate Selection

• Can be achieved using the roulette wheel
  algorithm.
• Construct a roulette wheel with a marker
• proportional to the fitness of each
• individual as shown.
• When the arrow is spun the
• probability of selecting an
• individual is thus propotional
• to the fitness of that individual.

11
Rank Selection

• All individuals are sorted according to their
  fitness.
• Each individual is then assigned a probability of
  being selected from some prior probability
  density.

12
Tournament Selection

• Select a group of N
• (Ngt1) members.
• Select the fittest member of this group and
  discard the rest.

13
Other Selection Techniques

• To overcome some of the problems associated with
  selection (e.g. stagnation and premature
  convergence), the following can be used
• Fitness scaling
• Ensures that extremely fit members are not
  selected too often during fitness proportionate
  selection methods.
• Elitism
• A small number of the best individuals are
  retained so that they will survive into the next
  generation.

14
Crossover

• The means by which individuals are combined to
  form offspring.

15
Mutation

• The Mutation operator ensures
• the gene pool does not become
• too restricted.
• In GAs it is carried out by randomly changing one
  or more of the alleles (bits) in an individuals
  chromosome.
• The probability of mutating a particular bit is
  typically very small ( 0.001).

16
Parallelising a Genetic Algorithm

• Genetic Algorithms are highly parallelisable
  since most of the operators can be caried out on
  individual members independently of other members.

17
Parallelisation Methods

• Common parallel GA prototypes
• Master Slave prototype.
• Distributed, Asychronous Concurrent prototype.
• Network model.
• Island model.

18
Master Slave prototype
19
Distributed, Asychronous Concurrent prototype.
20
Network Model
21
Island Model
22
Applications of GAs

• Optimisation tasks
• Automatic Programming
• Machine Learning
• Economics

• Immune Systems
• Ecology
• Population genetics
• Social Systems

23
Generic Parallel GA Function

• Population ParallelGeneticAlgorithm(
• int nislands, int ngenerations, int nmembers,
  int string_length,
• GA_Op select_type, double select_arg,
• int nelite, GA_Op cross_type,
• double cross_prob, int ncross_points,
• int gene_lengths, double mut_prob,
• GA_Op scaling_type, double scale_arg,
• GA_Op mig_type, double mig_prob,
• double (ObjectiveValueFunction)
• (Population , int, int, int)
• )