1 / 12

Chapter 3 GA2: Why do they work?

Chapter 3 GA2: Why do they work?. Schema don’t care symbol * 사용 예) chromosome length 10 (*111100100 matches {(0111100100), (1111100100)} (*1*11100100) matches {(01011100100), (11011100100), (01111100100), (11111100100)} r 개 *를 가진 string 은 2 r 개 string 과 match

archer
Download Presentation

Chapter 3 GA2: Why do they work?

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Chapter 3 GA2: Why do they work? • Schema • don’t care symbol * 사용 • 예) chromosome length 10 (*111100100 matches {(0111100100), (1111100100)} (*1*11100100) matches {(01011100100), (11011100100), (01111100100), (11111100100)} • r개 *를 가진 string은 2r개 string과 match • 길이 m인 string은 2m schema와 match

  2. Properties of schema • Order • o(S) = number of non-don’t care bits • 예) S1 = (***001*110), o(S1) = 6 S2 = (****00**0*), o(S2) = 3 S3 = (11101**001), o(S3) = 8 • used for defining the specialty of a schema • mutation에서 survival probability 계산에 사용 • Defining length • (S) = distance between the first and the last fixed string positions • 예) (S1) = 10-4 = 6, (S2) = 9-5=4, (S3) = 10-1 = 9 • define the compactness of information contained in a schema • crossover에서 survival probability 계산에 사용

  3. Reproductive schema growth equation • Schema의 다음 세대에서의 번식력 계산 • GA 알고리즘 t <- t+1 select P(t) from P(t-1) recombine P(t) // crossover and mutation evaluate P(t) • (S,t) • number of strings in a population at time t, matched by schema S • 예) Chapter 2의 예 사용 • pop_size=20, m=33 • v1~v20 (p.47, same as p.37) • S0 = (****111**************************)라면 • (S0,t)=3 // v13, v15, v16 matched by S0 • o(S0)=3, (S0)=2

  4. Reproductive schema growth equation • Fitness of a schema • eval(S,t) = average fitness of all strings in the population matched by S • assuming p strings {vi1, ….. vip}, • eval(S,t) = j=1p eval(vij)/p

  5. Selection 단계 reproductive schema growth equation • 하나의 string vi • selection probability pi=eval(vi)/F(t) • F(t)=total fitness of population at t • (S0,t+1)=(S,t)*pop_size*eval(S,t)/F(t) • (S0,t+1)=(S,t)*eval(S,t)/F(t) ------------ (3.1) • 특성 • “above average” schema: increasing in next generation • “below average” schema: decreasing in next generation • “average” schema: stay on the same level • long-term effect • “above average” schema increases exponentially in next generations

  6. Selection 단계 reproductive schema growth equation • 예) • S0 matches 3 strings v13, v15, v16 • eval(S0,t) = (27.316702+30.060205+23.867227)/3=27.081378 • F(t)= 387.776822/20=19.388841 • S0는 “above average” • eval(S0,t)/F(t) = 1.396751 • (S0,t+1)=3*1.396751=4.19 • (S0,t+2)=3*1.3967512=5.85 • …………… • Chapter 2에서는 • v1’~v20’ (p.39) • S0는 t+1에서 5개 string (v7’, v11’, v18’, v19’ v20’)과 match

  7. Crossover 단계 reproductive schema growth equation • 예) • v18’ matches S0=(****111***********…****) and S1=(111********…….*******10) • p.40의 crossover 예 • v18’=(11101111101000100011|0000001000110) • v13’=(00010100001001010100|1010111111011) • v18’’=(11101111101000100011|1010111111011) • v13’’=(00010100001001010100| 0000001000110) • S0는 survived • S1은 destroyed pos=20

  8. Crossover 단계 reproductive schema growth equation • Defining length가 survival/destruction에 중대한 영향 • probability of destruction • pd(S) = (S) /(m-1) • probability of survival • ps(S) = 1 - (S) /(m-1) • S0와 S1 예 • (S0)=2 --> pd(S0)=2/32, ps(S0)=30/32 • (S1) =32 --> pd(S1)=32/32, ps(S1)=0 • pc (crossover 확률) 고려하면 • ps(S) = 1 - pc*((S) /(m-1)) • 예) ps(S0) = 1 - 0.25*(2/32)=0.984375 • 우연히 survive할 경우 고려하면 • ps(S) >= 1 - pc*((S) /(m-1))

  9. Crossover 단계 reproductive schema growth equation • Selection (수식3.1)과 crossover 모두 고려한 수식 • (S,t+1) >= (S,t)*(eval(S,t)/F(t))*(1 - pc*((S) /(m-1))) ----- (3.2) • S0 예에 적용하면 • (eval(S,t)/F(t))*(1-pc*((S) /(m-1))) = 1.396751*0.984375=1.374927 • (S0,t+1) = 3*1.374927 = 4.12 • (S0,t+2) = 3*1.3749272 = 5.67 • ………….. • selection만 고려한 것보다 약간 작은 값

  10. Mutation 단계 reproductive schema growth equation • 예 • v19’=(111011101101110000100011111011110) • S0=(****111***************….******) • v19’에서 5,6,7번째 bit중 하나라도 mutate되면 S0는 destroyed • Mutation에 대한 survival 확률 • Ps(S)=(1-pm)o(S) • Ps(S)  1-o(S)*pm (since pm<<1) • 예) pm=0.01 • Ps(S0)  1-3*0.01 = 0.97

  11. Selection, crossover, mutation 모두 고려한 reproductive schema growth equation (S,t+1)>=(S,t)*(eval(S,t)/F(t))*(1-pc*((S) /(m-1))-o(S)* pm) ----- (3.3) • 예) • (eval(S,t)/F(t))*(1-pc*((S) /(m-1))-o(S)* pm) =1.396751*0.954375=1.333024 • (S0,t+1) = 3*1.333024 = 4.00 • (S0,t+2) = 3*1.3330242 = 5.33 • …………..

  12. Schema theorem & building block hypothesis • Schema theorem • “Short, low-order, above-average schemata receive exponentially increasing trials in subsequent generations of a genetic algorithm” • Building block hypothesis • “A genetic algorithm seeks near-optimal performance through the juxtaposition of short, low-order, high-performance schemata, called the building blocks”

More Related