Genetic Programming

EA

GP

GP
GP

GP
J. Koza

GP

Shape Size

GP GP

GP

GP
GP

GP

GP
 GP

GP

GP

GP
GP

GP
GP

GP GP

GP

GP

GP
 Reproduction

Crossover
GP

F={AND , = ,
>}
T={NOC ,2 ,S,80000}

Penalty

C= (FUT) AND

AND

NOC 2 S 80000

GP
GP

GP

GP
 y
2 ( x 3)
5 1

(x true) (( x y) (z (x y)))

i =1;
while (i < 20)
{
i = i +1
}

EP ES GA

GA GP
EP ES
GP

F T

t T
n f F f
f(e1..,..,en) en e1

arity
 GP
GP Mutation Crossover GP
ES GA
GA
GP

GA flowchart GP flowchart

GP
GP

GP

Reproduction
 Crossover

Reproduction
GA

n(i,t)

Rank Selection

Fitness Scaling
Tournament Selection

Crossover
GA GP GA GP Crossover

mating pool Mating pool

(Crossover) GP GA GP

GP Crossover

Crossover Crossover
Offspring
GA
Crossover
Crossover Crossover
Crossover GA

Crossover
GP GA Mutation

GP
 Parent 1 Parent 2

Child 1 Child 2
Crossover

GP
GP Reproduction , Crossover

Mutation
Permutation
Editing
Encapsulation
(Decimation)

Diversity
GP

Crossover

:
 GP

GA Inversion
K
k!

NOT(NOT(NOT(NOT(X))))=X X AND X X
Crossover GP
Editing Performance
f ed f ed Editing

Crossover

Decimation
Local
 GP
Over-Selection

100-x % x%

80%
20%
X%

x
32% 1000
16% 2000
8% 4000
4% 8000

Generational GP

Steady state

Full Method
Grow Method
Ramped half and half
 Full method

F Dmax
T Dmax

Dmax
Grow method
C F T
T

Dmax
Ramped half-and-half

Grow method Full method

Full method
Grow method

Diversity

Raw Fitness
Standardized Fitness
Adjusted Fitness
Normalized fitness

Raw Fitness
Raw Fitness
 Raw Fitness
Nc
Nc
r (i, t ) S (i , j ) C ( j )
j 1

C(j) j i S(i,j)
Boolean j

S(i,t)=r(i,t) Standardized Fitness

rmax S(i,t)= r(i,t)

Adjusted Fitness
1
a (i , t )
1 S (i, j )
a(i,t) S(i,t)

Normalized fitness

a (i, t )
n(i, t ) M
a (k , t )
k 1

GP
GP
GP
GP

GP GP

GP
GP
Reproduction

GP
 GP
(Major) GP
Minor

p c Crossover
p r Reproduction
Crossover
p ip
DCreated Crossover
Dinitial
pm
pp
f ed Editing
p en
True or False Decimation
pd Decimation

GP

Crossover

Elitist

GP Schema GP
GA
Building Crossover
Block

Symbolic Regression

Mean Square Root

Symbolic Regression
4 3 2
y x x x x
xi 1,1

S-expressions S-expression

x
T x x

x4 x3 x2 x

SIN ,COS,RLOG,EXP

F={+ , - , * , / , SIN , COS , EXP,RLOG

}

raw-fitness
yi xi S-expression
 GP
GP
G=51 GP

Standardized-fitness
raw-fitness

S-expression

S-expression
S-expression

GP

x4 x3 x2 x
S-expression 0.0
(+ X (* (+ X (* (* (+ X (- (COS(- X X )) (- X X))) X) X)) X))

COS
RLOG SIN EXP
 S-expression COS(- X X)
, SIN x4 x3 x2 x
RLOG EXP COS
This document was created with Win2PDF available at http://www.daneprairie.com.
The unregistered version of Win2PDF is for evaluation or non-commercial use only.