GAlib 2.4.2

[Matthew Wall]

Hello,

first of all, many apologies for letting the GADemeGA remain is such a
pathetic state for so long.  fixes are on their way...

>Hallo there!
>I'd like to have some information about this lib I am using, which
>from my point of view has quite some problems.
>
>Could anybody explain me what the following methods and properties
>for a DemeGA stand for:
>
>a. populationSize : the size of each population or the size of all
>                    populations? I understood it's the former but no
>                    information is given on the guide.
>
>b. nReplacement:   as it should be used to determine the number of
>                   individuals to be replaced, could you tell me why
>                   I get as a message: 'the nReplacement percentage
>                   should be kept between 0.0 and 1.0'? Isn't it this
>                   the range of values for pReplacement?
>
>c: Can the replacement and migration methods be used in conjunction or
>   the user has to choose one of them?

migration and replacement are independent.  migration controls how many
individuals move from one population to another when you are using a
multi-population genetic algorithm (e.g. GADemeGA)

pReplacement and nReplacement are not indepedent; you must specify one or
the other.  when you set nReplacement, the GA sets pReplacement based on
the nReplacement that you specified (and vice versa).  if the population
size(s) are messed up, then the GA will calculate a bogus pReplacement
based on your nReplacement, and the set on the parameter list will spit out
the '0.0 and 1.0' error message.  i believe this is fixed in the new
GADemeGA.C

>d. Why customized initialaters and/or other operators are not correctly
>   set for all members of the population?
>   I debug my code and I found out that *SOME* individuals had the
>   initializer set to SexCross and the mutator to AsexCross instead of
>   my functions!!!

i suspect that this is related to the other deme ga bugs.  the GADemeGA in
2.4.2 does not properly update its populations.

>e. Is there any memory restriction for the use of GAlib classes?
>   If not, could you explain me why the same program works fine with
>   a small population and it does not work (giving me the above mentioned
>   problems) with bigger ones?

there are significant bugs in this genetic algorithm - i suggest you use
galib 2.4.3.  the complete 2.4.3 release should be available within a day
or two.  i will send an announcement to galib-announce at mit.edu when it is
ready.

until then, you can grab the updated GADemeGA.C file from

ftp://lancet.mit.edu/private/mbwall/GADemeGA.C


matthew


Matthew Wall
mbwall_NO_SPAM at mit.edu
http://mit.edu/mbwall