Optimal Number of Optimal Number of Genders in EA for Genders in EA for Different Types of Different Types of Problems Problems CS448 Term Project CS448 Term Project Instructor: Daniel R. Tauritz Instructor: Daniel R. Tauritz Yin Liang Yin Liang November 28 November 28 th th , 2005 , 2005
24
Embed
Optimal Number of Genders in EA for Different Types of Problems
Optimal Number of Genders in EA for Different Types of Problems. CS448 Term Project Instructor: Daniel R. Tauritz Yin Liang November 28 th , 2005. Motivation. Multiple-genders are rarely explored and even experimented now Premature Convergence Equilibrium State Keep Balance. Main Goal. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Optimal Number of Genders in Optimal Number of Genders in EA for Different Types of EA for Different Types of
Problems Problems
CS448 Term ProjectCS448 Term Project
Instructor: Daniel R. TauritzInstructor: Daniel R. Tauritz
Yin LiangYin Liang
November 28November 28thth, 2005, 2005
MotivationMotivation
Multiple-genders are rarely explored Multiple-genders are rarely explored and even experimented nowand even experimented now
Premature ConvergencePremature Convergence
Equilibrium State Equilibrium State
Keep BalanceKeep Balance
Main GoalMain Goal
Different selection schemes for Different selection schemes for different gendersdifferent genders
Explore the optimal results using Explore the optimal results using multiple gendersmultiple genders
Intend to build a decision tree based Intend to build a decision tree based on the resultson the results
DesignDesign
No Specific RepresentationsNo Specific Representations
Initialize Population RandomlyInitialize Population Randomly
Para1: single gender (20-TS) Para1: single gender (20-TS)
Para 2: 2 genders(Random+20-TS)Para 2: 2 genders(Random+20-TS)
Para 1: get the optimal values afterPara 1: get the optimal values after 4347 4347 generations on averagegenerations on average
Para2: get the optimal values afterPara2: get the optimal values after 5011 5011 generations on averagegenerations on average
Para1: the maximum fitness through 30 Para1: the maximum fitness through 30 runs is 966runs is 966
Para2: the maximum fitness through 30 Para2: the maximum fitness through 30 runs is 968runs is 968
Statistically speaking, there is no Statistically speaking, there is no difference between Para1 and Para2 by difference between Para1 and Para2 by using t-test.using t-test.
Integer Vector (Assignment2)Integer Vector (Assignment2)Para 1: single gender (4-TS) Para 1: single gender (4-TS)
Para 2: 2 genders (Random+4-TS)Para 2: 2 genders (Random+4-TS)
Para 1: get the optimal values afterPara 1: get the optimal values after 57 57 generations on averagegenerations on average
Para2: get the optimal values afterPara2: get the optimal values after 87 87 generations on averagegenerations on average
Para1: the maximum fitness through 5 Para1: the maximum fitness through 5 runs is -0.411818runs is -0.411818
Para2: the maximum fitness through 5 Para2: the maximum fitness through 5 runs is -0.379094runs is -0.379094
Statistically speaking, there is no Statistically speaking, there is no difference between Para1 and Para2 by difference between Para1 and Para2 by using Wilcoxon rank sum test.using Wilcoxon rank sum test.
Different Population SizeDifferent Population SizePop Size = 50 Mutation Chance = 0.001Pop Size = 50 Mutation Chance = 0.001
Para 1: single gender (10-TS)Para 1: single gender (10-TS)
Para 2: 2 genders (Random+10-TS)Para 2: 2 genders (Random+10-TS)
Para 3: 3 genders (Random+FPS+10-TS)Para 3: 3 genders (Random+FPS+10-TS)
Statistically speaking, there is no Statistically speaking, there is no difference among Para1, Para2 and Para3 difference among Para1, Para2 and Para3 by using t-test.by using t-test.
Single gender is enough to produce the Single gender is enough to produce the optimal results in the population size of 50optimal results in the population size of 50
Para 1: single gender (80-TS)Para 1: single gender (80-TS)
Para 2: 2 genders (Random+80-TS)Para 2: 2 genders (Random+80-TS)
Para 3: 3 genders (Random+FPS+80-TS)Para 3: 3 genders (Random+FPS+80-TS)
Statistically speaking, there is no Statistically speaking, there is no difference among Para1, Para2 and Para3 difference among Para1, Para2 and Para3 by using t-test.by using t-test.
In 30 runs, the maximum fitness can be In 30 runs, the maximum fitness can be obtained by using 3 genders.obtained by using 3 genders.
Para 1: single gender (80-TS)Para 1: single gender (80-TS)
Para 2: 2 genders (Random+80-TS)Para 2: 2 genders (Random+80-TS)
Para 3: 3 genders (Random+FPS+80-TS)Para 3: 3 genders (Random+FPS+80-TS)
Statistically speaking, there is significant Statistically speaking, there is significant difference between Para1 and Para2 but difference between Para1 and Para2 but no difference between Para2 and Para3 by no difference between Para2 and Para3 by using t-test. using t-test.
In 30 runs, the maximum fitness can be In 30 runs, the maximum fitness can be obtained by using 2 or 3 genders.obtained by using 2 or 3 genders.
Different Mutation ChanceDifferent Mutation ChanceMutation Chance = 0.005 Population Size = 100Mutation Chance = 0.005 Population Size = 100
Para 1: single gender (40-TS)Para 1: single gender (40-TS)
Para 2: 2 genders (Random+40-TS)Para 2: 2 genders (Random+40-TS)
Para 3: 3 genders (Random+FPS+40-TS)Para 3: 3 genders (Random+FPS+40-TS)
Statistically speaking, there is significant Statistically speaking, there is significant difference among Para1, Para2 and Para3 difference among Para1, Para2 and Para3 by using t-test.by using t-test.
In 30 runs, the maximum fitness can be In 30 runs, the maximum fitness can be obtained by using 2 genders.obtained by using 2 genders.
Mutation Chance = 0.005 Population Size = 200Mutation Chance = 0.005 Population Size = 200
Para 1: single gender (80-TS)Para 1: single gender (80-TS)
Para 2: 2 genders (Random+80-TS)Para 2: 2 genders (Random+80-TS)
Para 3: 3 genders (Random+FPS+80-TS)Para 3: 3 genders (Random+FPS+80-TS)
Statistically speaking, there is significant Statistically speaking, there is significant difference among Para1, Para2 and Para3 difference among Para1, Para2 and Para3 by using t-test.by using t-test.
In 30 runs, the maximum fitness can be In 30 runs, the maximum fitness can be obtained by using 2 genders.obtained by using 2 genders.
ConclusionConclusion
Representations should not be included in Representations should not be included in choosing multiple genders.choosing multiple genders.
For population size less than 400, there is For population size less than 400, there is no difference by using multiple genders no difference by using multiple genders from the prospective of statistics. But the from the prospective of statistics. But the maximum fitness will obtain by using the maximum fitness will obtain by using the certain number of genders in the limited certain number of genders in the limited number of runs.number of runs.
ConclusionConclusion
For population size equal to or greater For population size equal to or greater than 400, using 2 or 3 genders will than 400, using 2 or 3 genders will improve the efficiency of EAs at the same improve the efficiency of EAs at the same level.level.
For the large mutation chance, 2 genders For the large mutation chance, 2 genders will be able to keep the balance between will be able to keep the balance between selection pressure and genetic diversity selection pressure and genetic diversity better than 1 or 3 genders.better than 1 or 3 genders.