problems with galib2.4.2
Ilkka Tapio Ikonen
itikon01 at starbase.spd.louisville.edu
Wed Feb 19 18:38:59 EST 1997
I am having some difficulties when switching from galib-2.4.0 to galib-2.4.2.
I have a GA with its own genome type (based on ex14) and it works fine with
galib-2.4.0. The new (2.4.2) ex14 doesn't evolve either, old one works fine.
New ex11 worked well. I read the notes about random number generators and
tested mine. Itpassed the tests nicely (many times) and change of random # generator makes
no difference to the GA. Currently I'm using RAN2 (and RAN3). I have WIN95
with Borland C++ 4.52.
I am not exactly sure what causes my problem (otherwise I would have fixed it
:-)
The GA evolves through generations, evaluates individual and calls and performs
crossovers and mutations. It improves the worst and average of the population,
but the best does not improve *at all*??? So, the best found is actually
the best in the 1st generation. Of course, the worst is very fast the same
as the best.
I am (still) suspecting that those random # problems would cause this, but
I just can't see how this could happen?? I looks more like the best is not
selected for crossover, or child from this xover is not put to the new
generation
(all ga types give the same problem), or scaling or selecting doesn't work
correctly (which
I don't think) or new population doesn't get evaluated and old evaluations
are kept instead??
One funny test:
As a genome evaluation I returned a GARandomInt(0,50) and still the same
thing. For each generation the best, average and the worst should be a
random number from this range. BUT: All are the same (straigh line instead
of scattered dots)?? And the average or the worst do not improve (or worsen)
at all.
I wonder what have I done now :-)
Ilkka
_____ _____
/ __) (__ \
/ (__|_________________________________________|__) \
/ | | \
/ | Ilkka Ikonen | \
\ | Ph.D. candidate, Graduate Teaching Assistant | /
\ | Department of Industrial Engineering | /
\ | University of Louisville | /
\__| T:+1-502-852 1695 |_ /
| Fax:+1-502-852 5633 |
| email: itikon01 at starbase.spd.louisville.edu |
| www: http://www.spd.louisville.edu/~itikon01 |
|_______________________________________________|
More information about the galib
mailing list