Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!news.starnet.net!wupost!nic.smsu.edu!newsdist.tc.umn.edu!umn.edu!news
From: "James Albert Larson" <larso171@maroon.tc.umn.edu>
Subject: Re: TSP programs for GAucsd
To: chenp@newton.ccs.tuns.ca
Message-ID: <23772.larso171@maroon.tc.umn.edu>
X-Minuet-Version: Minuet1.0_Beta_16
Sender: news@news.cis.umn.edu (Usenet News Administration)
Nntp-Posting-Host: dialup-2-153.gw.umn.edu
X-Popmail-Charset: English
Organization: University of Minnesota, Twin Cities
Date: Sun, 28 May 1995 08:09:48 GMT
Lines: 43

On Thu, 25 May 1995 17:39:11 GMT, 
Ping Chen  <chenp@newton.ccs.tuns.ca> wrote:

> Has anybody out there used GAucsd to solve anything useful, like the TSP?
> I have got the package set up and running, and now would like to get my
> hands on some actual problem that was solved using it. Please post replies
> either to this newsgroup or to ag562@leo.nmc.edu.

I haven't made the effort to try to run Genesis 1.2ucsd, but I have studied 
the user's guide.  It doesn't look like its set up to solve the TSP, 
without a lot of additional user code.  To solve TSP with GA, one 
must come up with special order-based encodings, e.g. ( 1 3 4 2 5 ) 
describes a tour 1->3->4->2->5->1, and special mutation and crossover 
operators that work on this kind of encoding, and produce feasible tours.  
Even then, blind GA will solve only relatively small TSP problems (< 100 
cities).

There are a number of different encodings (I just showed the conceptually 
simplest above), but all of the ones (talking off the top of my 
head) described in Michalewicz's book, "Genetic Algorithms + Data 
Structures = Evolution Programs" 1994, are similar to the one I showed, in 
that they all involve a list of city numbers.

To solve larger TSP problems, one must use some heuristic information and 
operators.  E.g. when given a choice of edges to extend a tour,  pick the 
shortest edge.

While on the subject of Genesis 1.2ucsd, has anyone gotten it to run on a 
PC?  It looks like its pretty customized for Unix, and I haven't taken the 
time to study the code to see what it would take to run it on a DOS PC.

And while rambling on and wasting bandwidth, I've gotten Genocop3.0.tar.z 
to run on a PC, compiling with Borland C++ 4.0, in case anyone is 
interested.  Its described in the Michalewicz book mentioned above.  Its a 
GA written in C, customized to solving optimization problems involving 
nonlinear objective functions and linear constraints.  (He's got a version 
now posted on  ftp.uncc.edu  directory coe/evol  for solving problems 
involving nonlinear constraints, but I haven't studied it).  (He's also 
produced a readme file recently, a few weeks ago, that describes what's in 
the directory).

Jim Larson

