GENETIC ALGORITHMS

by

Volkan TUNALI
Content
 Evolutionary Computation (EC)
 Positioning of EC
 Some terminology
 Problem types
 Evolutionary Algorithms (EA)
 Genetic Algorithms (GA)
 Original GA by Holland
 Example
 Sample Application – Travelling Salesman
Positioning of EC
 EC is part of computer science

 EC is not part of life sciences/biology

 Biology delivered inspiration and

terminology

 EC can be applied in biological research

Main EC Metaphor
EVOLUTION PROBLEM SOLVING

Environment Problem
Individual Candidate Solution
Fitness Quality

 Fitness  chance for survival and reproduction

 Quality  chance for seeding new solutions
Darwinian Evolution 1:
Survival of the Fittest
 All environments have finite resources
 (i.e., can only support a limited number of
individuals)

 Some kind of selection is inevitable

 Those individuals that compete for the

resources most effectively have
increased the chance of reproduction
Darwinian Evolution 2:
Diversity Drives Change
 Phenotypic traits
 Behaviour / physical differences that affect response to
environment
 Partly by inheritance, partly by factors during development
 Unique to each individual, partly result of random changes

 If phenotypic traits:
 Lead to higher chances of reproduction
 Can be inherited

Then they will tend to increase in subsequent generations

 leading to new combinations of traits..

Biological Terminology [1]
 Population: a group of individuals that may interact
with each other and occupy a given area at the same
time
 Adaptation: the evolution of features that make a
group of organisms better suited to live and reproduce
in their environment
 Fitness: the relative ability to leave offspring and
survive
 Natural Selection: a process of interaction between
organisms and their environment that results in a
differential rate of reproduction of different phenotypes
in a population (survival of the fittest)
 Inheritance: propagation of genetic material from
parent to offspring
Biological Terminology [2]
 Chromose: one of the bodies in the cell nucleus, along
which the genes are located (“blueprint” for the
organism)
 Gene: a functional block of DNA which encodes a
particular protein
 Genotype: the genetic constitution of a cell or and
organism
 Phenotype: observable properties of an organism
 Mutation: the inheritable change of a gene from one
allelic form to another
 Recombination: the formation of gene combinations
through mixing that differ from the combinations present
in the parents
 Crossing over: exchange of genetic material between
paired chromosomes
EAs  Real World Terminology
 Individual: one point in the search space of all
possible solutions
 Population: whole of the solution points currently
under consideration
 Gene: an encoding of a parameter or a part of a
solution
 Chromosome: an encoding of all the parameters
or the whole current solution candidate
 Crossing over: a recombinations operator used
during reproduction phase
 Mutation: abrupt change in gene value
EAs  Real World Terminology
 Phenotype: objects forming the possible
solutions within the original problem context
 Genotype: encoding of the phenotype that is
an individual within the EA
 Ex: Given an optimization problem on integers
 Given set of integers form the set of phenotypes
 18 is a phenotype & 10010 is a genotype
representing it.
 Fitness: how much the current solution meets
the requirements of the objective function
 Generation: one loop of the genetic algorithm
Problem Type 1: Optimization
 We have a model of our system and seek inputs
that give us a specified goal
 e.g. time tables for university, call center, or
hospital
Optimisation Example:
University Timetabling
 Enormously big search space

 Timetables must be good

 “Good” is defined by a number of competing

criteria

 Timetables must be feasible

 Vast majority of search space is infeasible

Problem Type 2: Modelling
 We have corresponding sets of inputs &
outputs and seek model that delivers correct
output for every known input
Modelling Example:
Loan Applicant Creditibility
 British bank evolved creditability
model to predict loan paying behavior
of new applicants

 Evolving: prediction models

 Fitness: model accuracy on historical

data
Problem Type 3: Simulation
 We have a given model and wish to know the outputs
that arise under different input conditions

 Often used to answer “what-if” questions in evolving dynamic environments

 e.g. Evolutionary economics, Artificial Life
What is an Evolutionary Algorithm?
 EAs fall into the category of “generate
and test” algorithms
 They are stochastic, population-based
algorithms
 Variation operators (recombination and
mutation) create the necessary diversity
and thereby facilitate novelty
 Selection reduces diversity and acts as a
force pushing quality
General Scheme of EAs
Pseudo-code for Typical EA
What are the Different Types of EAs
 Historically different flavours of EAs have been
associated with different representations
 Binary strings : Genetic Algorithms
 Real-valued vectors : Evolution Strategies
 Finite state Machines: Evolutionary Programming
 LISP trees: Genetic Programming
 These differences are largely irrelevant, best
strategy
 choose representation to suit problem
 choose variation operators to suit representation
 Selection operators only use fitness and so are
independent of representation
Representations
 Candidate solutions (individuals) exist in phenotype space
 They are encoded in chromosomes, which exist in genotype space
 Encoding : phenotype=> genotype (not necessarily one to one)
 Decoding : genotype=> phenotype (must be one to one)
 Chromosomes contain genes, which are in (usually fixed)
positions called loci (sing. locus) and have a value (allele)

In order to find the global optimum, every feasible solution

must be represented in genotype space
Evaluation (Fitness) Function
 Represents the requirements that the population
should adapt to
 a.k.a. quality function or objective function
 Assigns a single real-valued fitness to each
phenotype which forms the basis for selection
 So the more discrimination (different values) the

better
 Typically we talk about fitness being maximised
 Some problems may be best posed as

minimisation problems, but conversion is trivial

Population
 Holds (representations of) possible solutions
 Usually has a fixed size and is a multiset of
genotypes
 Some sophisticated EAs also assert a spatial
structure on the population (e.g. distance
measure or a neighborhood relation)
 Selection operators usually take whole
population into account i.e., reproductive
probabilities are relative to current generation
 Diversity of a population refers to the number of
different fitnesses / phenotypes / genotypes
present (**note not the same thing)
Parent Selection Mechanism
 Assigns variable probabilities of individuals acting as
parents depending on their fitnesses
 Usually probabilistic
 high quality solutions more likely to become

parents than low quality

 but not guaranteed

 even worst in current population usually has non-

zero probability of becoming a parent

 This stochastic nature can aid escape from local
optima
Variation Operators
 Role is to generate new candidate solutions
 Usually divided into two types according to
their arity (number of inputs):
 Arity 1 : mutation operators
 Arity >1 : Recombination operators
 Arity = 2 typically called crossover
 There has been much debate about relative
importance of recombination and mutation
 Nowadays most EAs use both
 Choice of particular variation operators is
representation dependant
Mutation
 Acts on one genotype and delivers
another
 Element of randomness is essential and
differentiates it from other unary
heuristic operators
 Importance depends on representation
and dialect
 May guarantee connectedness of search
space
Recombination
 Merges information from parents into
offspring
 Choice of what information to merge is
stochastic
 Most offspring may be worse, or the same
as the parents
 Hope is that some are better by
combining elements of genotypes that
lead to good traits
Survivor Selection
 a.k.a. replacement
 Most EAs use fixed population size so
need a way of going from (parents +
offspring) to next generation
 Often deterministic
 Fitness based : e.g., rank

parents+offspring and take best

 Age based: make as many offspring

as parents and delete all parents

 Sometimes do combination (elitism)
Initialisation / Termination
 Initialisation usually done at random,
 Need to ensure even spread and mixture of possible allele values
 Can include existing solutions, or use problem-specific heuristics,
to “seed” the population

 Termination condition checked every generation

 Reaching some (known/hoped for) fitness
 Reaching some maximum allowed number of generations
 Reaching some minimum level of diversity
 Reaching some specified number of generations without fitness
improvement
Example: 8 Queens Problem

Place 8 queens on an 8x8 chessboard in

such a way that they cannot check each other
 8 Queens Problem: Representation

Phenotype:
a board configuration

Genotype: Obvious mapping

a permutation of
the numbers 1 - 8 1 3 5 2 6 4 7 8
8 Queens Problem: Fitness
Evaluation
 Quality q(p) of any phenotype p
 the number of queens she can check

 q(p) = 0 indicates a good solution

 Objective function to be minimized

 Fitness of a configuration:
 inverse of q(p) (= 1 / q(p))
 M - q(p) where M is sufficiently large number
8 Queens Problem: Mutation
Small variation in one permutation, e.g.:
swapping values of two randomly chosen positions,

1 3 5 2 6 4 7 8 1 3 7 2 6 4 5 8
8 Queens Problem: Recombination
Combining two permutations into two new permutations:
 choose random crossover point
 copy first parts into children
 create second part by inserting values from other
parent:
 in the order they appear there
 beginning after crossover point
 skipping values already in child

1 3 5 2 6 4 7 8 1 3 5 4 2 8 7 6
8 7 6 5 4 3 2 1 8 7 6 2 4 1 3 5
8 Queens Problem: Selection

 Parent selection:
 Pick 5 parents and take best two to
undergo crossover
 Survivor selection (replacement)
 Merge the population and offspring
 Rank them according to fitness
 Delete the worst two
8 Queens Problem: Summary

Note that is only one possible

set of choices of operators and parameters
 Typical Behaviour of an EA

Phases in optimising on a 1-dimensional fitness landscape

Early phase:
quasi-random population distribution

Mid-phase:
population arranged around/on hills

Late phase:
population concentrated on high hills
 Typical Run: Progression of Fitness
Best fitness in population

Time (number of generations)

Typical run of an EA shows so-called “anytime behavior”

Are Long Runs Beneficial?
Best fitness in population

Progress in 2nd half

Progress in 1st half

Time (number of generations)

• Answer:
- it depends how much you want the last bit of progress
- it may be better to do more shorter runs
 Is It Worth Expending Effort on
Smart Initialisation?
Best fitness in population

F F: fitness after smart initialisation

T: time needed to reach level F after random initialisation

T
Time (number of generations)
• Answer : it depends:
- possibly, if good solutions/methods exist
Genetic Algorithms
 Developed: USA in the 1970’s
 Early names: J. Holland, K. DeJong, D. Goldberg
 Typically applied to:
 discrete optimization
 Attributed features:
 not too fast
 good heuristic for combinatorial problems
 Special Features:
 Traditionally emphasizes combining information from
good parents (crossover)
 many variants, e.g., reproduction models, operators
Genetic Algorithms
 Holland’s original GA is now known as the
simple genetic algorithm (SGA)
 Other GAs use different:
 Representations
 Mutations
 Crossovers
 Selection mechanisms
SGA Technical Summary
Representation Binary strings
Recombination N-point or uniform
Mutation Bitwise bit-flipping
with fixed probability

Parent selection Fitness-Proportionate

Survivor selection All children replace

parents
Speciality Emphasis on
 Representation

Phenotype space Genotype space =

Encoding {0,1}L
(representation) 10010001
10010010
010001001
011101001
Decoding
(inverse representation)
SGA Reproduction Cycle

1. Select parents for the mating pool

(size of mating pool = population size)
2. Shuffle the mating pool
3. For each consecutive pair apply crossover with
probability pc , otherwise copy parents
4. For each offspring apply mutation (bit-flip with
probability pm independently for each bit)
5. Replace the whole population with the resulting
offspring
SGA Operators: 1-point Crossover
 Choose a random point on the two parents
 Split parents at this crossover point
 Create children by exchanging tails
 Pc typically in range (0.6, 0.9)
SGA Operators: Mutation
 Alter each gene independently with a probability pm
 pm is called the mutation rate
 Typically between 1/pop_size and 1/ chromosome_length
SGA Operators: Selection
 Main idea: better individuals get higher chance
 Chances proportional to fitness
 Implementation: roulette wheel technique
 Assign to each individual a part of the roulette wheel
 Spin the wheel n times to select n individuals

1/6 = 17%

A B fitness(A) = 3
C fitness(B) = 1
3/6 = 50% 2/6 = 33%
fitness(C) = 2
An Example After Goldberg ‘89
 Simple problem: max x2 over {0,1,…,31}
 GA approach:
 Representation: binary code, e.g. 01101 ↔ 13
 Population size: 4
 1-point xover, bitwise mutation
 Roulette wheel selection
 Random initialisation
 We show one generational cycle done by hand
x2 Example: Selection
x2 Example: Crossover
x2 Example: Mutation
The Simple GA
 Has been subject of many (early) studies
 Still often used as benchmark for novel GAs
 Shows many shortcomings, e.g.
 Representation is too restrictive
 Mutation & crossovers only applicable for bit-
string & integer representations
 Selection mechanism sensitive for converging
populations with close fitness values
 Generational population model (step 5 in SGA
repr. cycle) can be improved with explicit
survivor selection
Conclusion with a Sample
Application
 Travelling Salesman Problem
 visit a number of cities once
 minimize the overall distance travelled
 As the number of cities grow, the time
required to find the solution grows
exponentially = O(2n)
 See the demonstration...