Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!torn!news!spicer
From: spicer@server.uwindsor.ca (Spicer Andrew)
Subject: Creating the First Generation
Message-ID: <D0Gqn5.D4I@uwindsor.ca>
Sender: news@uwindsor.ca (Usenet)
Organization: University of Windsor, Ontario, Canada
Date: Wed, 7 Dec 1994 23:01:52 GMT
Lines: 42

I am trying to use genetic algorithms to solve a particular
problem, and it seems to work for small problems.

When I get to bigger problems, I have troubles, though.
It seems that it is quite likely that I can have a beginning
generation that is made up entirely of infeasible solutions.
This screws up my process for selecting parents for the
crossovers as the whole generation has fitness=0.

I have tried rejecting any infeasible solution for inclusion
in the first generation, but the program seems to randomly
search through 100's of thousands of solutions before finding
feasible ones.

Here are my questions:

1)  What suggestions do you have about creating my first
generation?

2)  If my feasible solutions are so few and far between,
will I have more trouble when I start crossing over?  What
can be done in this case?

3)  When I say that my program seems to work for small 
problems, I am not sure if it really is okay.  Sometimes
it finds the optimal solution, other times it doesn't --
this I expect...  But whatever answer it gives me is
usually found in the first 100 generations.  After that
I don't get much progress.  (i.e., my optimal fitness
is 780, and after 50 generations I have a best-so-far of
650, and then I am stuck there...)

Thanks in advance for any advice,

Andrew Spicer
University of Windsor


-- 
-----------------------------------------------------------------------
  Andrew Spicer  University of Windsor  spicer@cobra.esxf.uwindsor.ca
 <A HREF="http://cobra.esxf.uwindsor.ca/spicer.html">Andrew Spicer</A>
