Genetic Algorithm By Eric Hilton http:compsci'snc'educs_460hiltep

1
Genetic AlgorithmBy Eric Hiltonhttp/compsci.s
nc.edu/cs_460/hiltep
2
Genetic Algorithm

• A technique used to help solve optimization
  problems
• Does not grantee optimal solutions The goal is
  that it will get better - positive slope
• Is very interesting in other fields of study
  particularly Economics and Social Sciences And
  of course Evolutionary Sciences
• Schedulers
• Kind of fun?

3
History

• As old as Time?
• Began in 1954 by Nils Aall Barricelli at
  Princeton simply as a way of studying computer
  simulations of evolution.
• Others followed slowly adapting different
  changes.
• In the 1960s-1970s many people began to realize
  the Algorithm may have use in the field of
  Optimization

4
History Continued

• With the expanding capability of hardware in the
  1980s there became possibilities for the GA to
  be implemented.
• The first known GA product was marketed by
  General Electric, a mainframe-based toolkit
  designed for industrial processes.
• In 1989 Axcelis, Inc released the Evolver the
  second known GA product and the first for
  Personal Computers.

5
Problem

• Use and Analyze the Genetic Algorithm to solve a
  problem.
• The chosen problem is a shortest distance between
  two points problem.

6
Problem Continued

• Shortest distance is by going from vertex to
  vertex.
• All shapes must be convex.
• Start and end are not fixed places.

7
The Double Helix Revisited

• Rules of the game
• To represent a path to the computer it must be a
  string of desired moves from one place to the
  next.
• Everyone starts at start or position 0.
• Everyone moves until
• they make an illegal move and they die
• until they win
• n moves have been made
• In the picture n14

8
Double Helix Revisited

• The pink doted lines represent illegal moves from
  the last stop that the yellow line made, on the
  bottom
• Illegal one would be 0,1,6
• A legal winning path on the top diagram would be
  0,2,4,13,12,6,14

9
How Do They Connect?

• One person try 50 times?
• 50 people try 1 time?
• Learning from each trial.
• Rules of how to learn.
• Elitism?
• Survival of the fittest mostly

10
Rules of Survival

• Generate a population of size P.
• Each chromosome in Population P has a randomly
  generated string of moves.
• Run the simulation i.e. let each chromosome try
  to get to the goal
• Rate each chromosome on its progress.
• Possibly cross over the selected chromosomes
• Possibly mutate the selected chromosomes

11
Rating Chromosomes

• Each time a Chromosome tries to get to its goal
  it gets a rating on how well it achieved this
  task.
• Formula for rating
• Note
• AM Accomplished Moves
• TP Total Possible Moves
• If the chromosome died from an illgeal move or
  reached n moves but not the goal
• Rating AM
• If the chromosome found the goal legally
• Rating (TP) (TP AM)

12
Roulette Wheel Selection

• Imagine spinning a roulette wheel
• But your fitness rating gives you a relative
  proportion of the wheel rather than each section
  being the same size.
• If youre number is picked you are able to
  procreate sexual/asexual and mutation

13
Sexual Procreation Crossing Over

• Cross over point is 6
• 0 10 4 8 6 12 10
• 0 3 8 9 10 7 12
• 0 10 4 8 6 10 7 12
• 0 3 8 9 6 12 10

14
Mutation

• Runs through the string of possible moves and
  selects certain ones to be mutated
• 0 10 4 8 6 12 10
• 0 2 4 8 6 14 10

15
Neither?

• There is the chance that neither would happen.
• In this instance there would be an asexual
  procreation
• In other words the exact same chromosome would go
  into the next generations population

16
Run again and again and again

• Continues to run in hopes of getting better and
  better.
• Most often does through my implementation

17
Methodology

• Adapt a working model I found to my problem.
• Implement it.
• The possible moves are represented in an
  unnecessarily large matrix array TPTP of
  legal and illegal moves.
• Population is stored in an array PTP
• Fitness is stored in an array P

18
Strategies

• Learn your problem inside and out before
  starting Research!!!
• Use constants!
• Good OOP is helpful
• Function use
• Local variables
• Private and public members