an analysis of edge assembly crossover for the traveling salesman problem yuichi nagata and...

33
An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation, 1999

Upload: roland-barker

Post on 17-Jan-2016

285 views

Category:

Documents


4 download

TRANSCRIPT

Page 1: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

An Analysis of

Edge Assembly Crossover for the Traveling Salesman Problem

Yuichi Nagata and Shigenobu Kobayashi

IEEE, Conference on Evolutionary Computation, 1999

Page 2: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Genetic AlgorithmHolland, 1975 -

Imitation of Evolution of life form in the Natureindividuals - members of species - in the nature

model of evolution processes where the basic operations are natural selection, crossover and mutationSchema Theorem - analysis of reproduction model

Nothing to do with the real genetic organism

Page 3: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Problem SolvingPolynomial time function

y = ax2 + b• Given constant a, b, know x, calculate y• Find x that gives Maximum y in a given range of x

No Calculation Search for the Solution

Search Space• For small space, use classical exhaustive techniques • For larger space, need special techniques• Analysis of space• Global Optima vs Local Optima

Page 4: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Stochastic SearchLocal Search TechniquesAdaptive Search Techniques

Random searchAnt ColonySimulated AnnealingNeural Network

search space

Page 5: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Standard Genetic Algorithm

Procedure GAbegin

t := 0 ;initialize P(t) ;evaluate P(t) ;while (not termination-condition) dobegin

t := t + 1 ;select P(t) from P(t-1) ;alter P(t) ;evaluate P(t) ;

end

end

Step 0: InitializationStep 0: Initialization

Step 1: SelectionStep 1: Selection

Step 2 : CrossoverStep 2 : Crossover

Step 3 : MutationStep 3 : Mutation

Step 5 : Termination TestStep 5 : Termination Test

Step6: EndStep6: End

Step 4: EvaluationStep 4: Evaluation

Page 6: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

GAs: Terminology

– Representation : gene, chromosome, Population– Evaluation : objective function, fitness function– Selection – Operator : crossover, mutation– Replacement : new Generation– Termination : Generation count, Convergence

Step 0: InitializationStep 0: Initialization

Step 1: SelectionStep 1: Selection

Step 2 : CrossoverStep 2 : Crossover

Step 3 : MutationStep 3 : Mutation

Step 5 : Termination TestStep 5 : Termination Test

Step6: EndStep6: End

Step 4: EvaluationStep 4: Evaluation

Page 7: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Representation

• Very crucial step• representation should satisfy the presumption that the whole

chromosome is decomposable to building blocks • String of genes of given alphabet:

– Binary – Float– Integer

• More complex representation– matrices– rules– trees

Page 8: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Initialization of the Starting Population

• Aspects affecting a performance of GA– schemata sampled in the initial population

• Initialization mechanisms – random – informed - uses prior knowledge of the desired solution shape

• Pre-processing– runs several short pre-processing runs– samples the promising areas of the search space identified

during the foregone pre-processing runs

Page 9: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Selection

• Models nature’s survival-of-the-fittest principle• Selection strategies:

– Roulette wheel (proportionate)– Ranking– Tournament

• Selection process:

– determination of Expected values: EVi = fitnessi / fitnessavg

– sampling algorithm - conversion of EVi to the actual number of individuals

Page 10: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Roulette Wheel Selection

Page 11: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Crossover

• Provides random information exchange - works on couples of individuals

• Simple 1-point crossover

Page 12: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Mutation

• Mutation - preserves population diversity– works on single individual

Page 13: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Replacement Strategy• Replacement strategy defines:

– how big portion of the old population will be replaced in each generation of the new population, and

– the rule that determines which individuals from the old population will be replaced and which individuals will be placed in the new population

• Generational - the old population is entirely rebuilt in each generation (short-lived species)

• Steady-state - just a few individuals are replaced in each generation (longer-lived species)

Page 14: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Premature Convergence

• The ratio of the best-fit individual’s reproduction rate to the average reproduction rate is too high

• selection kills ‘worse’ individuals too early

Page 15: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Theory

of GAs

Page 16: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Schema Theorem

• Schema = Pattern• Schema Theorem

– Short, low-order, above-average schemata – receive exponentially increasing trials in subsequent generations

of a genetic algorithm

• Building Block Hypothesis – GA seeks near-optimal performance through short, low-order,

high-performance schemata

mc pso

m

SptFtSevaltStS )(

1

)(1)(/),(),()1,(

Page 17: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

• Schema In binary representation - 2L strings, 3L schemata

L = 7, S = (**0*1*1) - covers 24 strings– {0,1, *}– S = {*1*01***, 1*0*11*0, 10111011, *******1, ****0*** }

• Fitness of a schema - average fitness computed over all covered strings

• Schema property– order

• the length of string minus the number of *• defining the specialty of a schema• 8 bits : 11010011, schema and building block 1*010*1*

– defining length • the distance between the first and the last fixed string positions• defining the compactness of information contained in a schema (*11**1*0) = 6, (1******1) = 7

)(So

)(S

Page 18: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Selection• eval(S,t) is the average fitness of all strings in the

population matched by the schema S at time t ;

• Expecting to have strings matched by schema S

– the average fitness of the population

– becomes ;

p

j j pvevaltSeval1

/)(),(

)(/),(|)(|),()1,( tFtSevaltptStS

|)(|/)()( tptFtF

)(/),(),()1,( tFtSevaltStS

)1,( tS

Page 19: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Crossover

)10****************************111(

*)*************************111***(*

1

0

S

S

)011101011111110010101000001010000(

,)110000000100010001000111110111110(

b

a

v

v

)110000000100010010101000001010000(

,)011101011111110001000111110111110(

b

a

v

v

– the string is matched by these two schemata

survives

destroyed

Page 20: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

– the probability of destruction of a schema S :

– the probability of survival of a schema S :

1

)()(

m

SSpd

1

)(1)(

m

SSps

(S) = 7, m = 8 ?

mc pso

m

SptFtSevaltStS )(

1

)(1)(/),(),()1,(

Page 21: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

• Selective probability of crossover

• The combined effect of selection and crossover

• a new schema growth equation :

cp

1

)(1)(

m

SpSp cs

1

)(1)(/),(),()1,(

m

SptFtSevaltStS c

Page 22: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

• All of the fixed positions of a schema must remain unchanged to survive mutation

• mutate at least one of these bits would destroy the schema

• the probability of destruction of a schema S :

• the probability of survival of a schema S :

)110001111101101110000101110111011(av

mpso )(

mpso1 )(

mc pso

m

SptFtSevaltStS )(

1

)(1)(/),(),()1,(

Mutation

Page 23: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

TSP with GA

Page 24: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Path representation

(5 1 7 8 9 4 6 2 3) 5-1-7-8-9-4-6-2-3

• Crossover operators Node Orientation vs Edge Orientation

• Mutation operators–insertion 5-2-1-7-8-9-4-6-2-3–Reciprocal Exchange 5-9-7-8-1-4-6-2-3–Inversion 5-9-8-7-1-4-6-2-3

Information of the parents transferred to offsprings

Node crossover = simple but information discarded

Edge crossover = tough but information enclosed

Page 25: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

TSP: Edge-Recombination Operator

b-c-e-a-d

b-d-e-c-a

a-b-c-d-e

Page 26: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Edge Assembly Crossover

Page 27: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Edge Assembly Crossover

Page 28: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

Previous work

Exx crossover

Ex crossover

Test Library

Page 29: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

EXXCrossover

Page 30: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

EXCrossover

Page 31: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

EXXCrossover

Page 32: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,

att532

Page 33: An Analysis of Edge Assembly Crossover for the Traveling Salesman Problem Yuichi Nagata and Shigenobu Kobayashi IEEE, Conference on Evolutionary Computation,