Population does not change GARealGenome

Murat Saginc koolaid2000man at yahoo.com
Fri Apr 12 06:39:41 EDT 2002

The selection of the parameters i think this quote says alot about:
"Selecting the various genetic algorithm parameters, such as mutation rate and type of crossover, is still more of an art than a science (Haupt and Haupt 1998, 64)." 
The following is from: Melanie Mitchell's "An Introduction to Genetic Algorithms"
(keep in mind that these numbers dont necessarily provide optimal results for all fitness functions.)
(have you checked out galib examples that use GARealGenome?)
1) De Jong's earlier studies:
pop_size=50-100; best single_pt crossover=0.06; pmut=0.001
On-line: psize=30; pcross=0.95; pmut=0.01; gen_gap=1; elitist selection
Off-line: De Jong and Grefensetter perform more or less the same
3)Schaffer, Caruana, Eshelman, & Das (1989):
On-line: psize=20-30; pcross=0.75-0.95; pmut=0.005-0.01
4) Self-adaptation
hehe goodluck there
side note:  the following is a good param setting used by Haupt and Haupt (1998, 49-65) for continuous parameter ga strictly in their example:
           (Ngood+Nbad=Npop) So every generation Nbad is dropped to be replaced by the new offspring from Ngood parents. Perhaps use 50% good/ 50% bad ratio. this is upto you how u do ur selection.
   pmut=0.04(1 to 20% is the range you should test in)
  crossover in cont => Adewuya(1996) & Michalewicz (1994) review the methods for doing continuous mating
                            => Radcliff, 1991 defines this as:
                                                 Pnew=B Pmn + (1-B)Pdn where
                                                  B = random number [0,1]
                                                  Pmn=the nth parameter in the mother chromosome
                                                  Pdn=the nth parameter in the father chromosome
                                                  (the second term in the eqn is the complement of the 1st)
 B = 0.5(Davis, 1991)=>avg. of two parents (Michalewicz, 1994) shows that this works well on bunch of problems
The simplest is linear crossover (Wright, 1991).  Three offspring from two parents are produced by:
       Pnew1 = 0.5Pmn + 0.5Pdn
       Pnew2 = 1.5Pmn - 0.5Pdn
       Pnew3 = -0.5Pmn + 1.5Pdn
If you did heuristic(probabilistic) crossover (Michalewicz, 1991) B is chosen random from [0,1] and
      Pnew = B(Pmn - Pdn) + Pmn
There are also other ones such as BLX-alpha (Eshelman & Shaffer, 1993) or Quadratic crossover (Adewuya, 1996
 With GALIB,  uniform initializer, gaussian mutator, uniform crossover are defaults
ex 21 & 23
if u want low gen & pop than u will use higher crossover for sure and mutation u would have to play with.
murat saginc
Computer Science
The College of Wooster
Wooster, OH
  Naveeed Ahmad <noni_4444 at hotmail.com> wrote: Hi all,
i am using galib to optimize the shape of fuzzy membership functions of a 
fuzzy ball beam system.
The problem i face is that my population does not change after i callthe 
evolve function. i am using GARealGenome with GASimpleGA. i cant run the 
program for long time, so i have to keep the number of generations and 
population size minimum.
Currently i have a popluation size of 4 and 10 number of generations...It 
already takes about 15 minutes to find a solution(which is similar to the 
string initially fed to the GARealGenome).
Can anyone suggest minimum number of generations and populations to observe 
some change. Should i change any other parameters..
Thanks in advance
Naveed Ahmad
National University of Computer and Emerging Sciences
Lahore Pakistan

Email: 972 at nu.edu.pk
noni_4444 at hotmail.com

Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp.

galib mailing list
galib at mit.edu

Do You Yahoo!?
Yahoo! Tax Center - online filing with TurboTax
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/pipermail/galib/attachments/20020412/b0dc5077/attachment.htm

More information about the galib mailing list