genetic algorithms by: anna scheuler and aaron smittle

15
Genetic Algorithms By: Anna Scheuler and Aaron Smittle

Upload: kenzie-buffkin

Post on 15-Dec-2015

213 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

Genetic Algorithms

By: Anna Scheuler and Aaron Smittle

Page 2: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

● appeared in the 1950s and 1960s● used to find approximations in search

problems● use principles of natural selection to find

an optimized solution● part of evolutionary algorithms

What is it?

Page 3: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

• subset of evolutionary computation

• generic, population based optimization algorithms

• uses aspects of biology

Evolutionary Algorithms

Page 4: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

• Gene = smallest unit of datao represented in binary

• Genome = string of genes

• Genome pool = set of genomeso represents the population

• Mutation

• Crossover

• Inheritance

Biology → Genetic Algorithms

Page 5: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

• Loops through every gene of every member

• Two main classes:o no changeo mutable

The Fitness Function

Page 6: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

1. Randomly generate an initial population

2. Run fitness function

3. Define parameters for “strong” members

4. Create new generation

5. Introduce mutation

6. Repeat

A simple algorithm runs in O(g*n*m)

The Algorithm

Page 7: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

• Opponent adaptation

• Towers of Reus

GAs and Gaming

Page 8: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

• Created in 2010 for Zerg

• user inputs goal and the app generates the build order

Star Craft’s Evolution Chamber

Page 9: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

● There are 10 cards numbered 1-10.● There must be two piles

○ The sum of the first pile must be as close as possible to 36

○ The product of the second pile must be as close as possible to 360

Card Problem Example

Page 10: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

● Genome is the way the cards are divided● Algorithm begins by picking two genomes

at random● They are compared with Fitness test● Copy winner into loser and mutate with

random probability at each gene

Card Problem cont.

Page 11: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

Card Problem Fitness Function

Page 12: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

● This problem used a Microbial GA○ This type of genetic algorithm features ‘free’

elitism○ Relatively simple core code

Card Problem

Page 13: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

http://rednuht.org/genetic_cars_2/

An example

Page 14: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

• The fitness function must be carefully written

• Members can get lost

• Population can converge with similar traits

Issues

Page 15: Genetic Algorithms By: Anna Scheuler and Aaron Smittle

Questions?