Re: GAs and TSP-Type Problems

On Thu, 20 Nov 2003 16:58:07 +0000 (UTC), "Kent Paul Dolan"
<[EMAIL PROTECTED]> wrote:

>"Steve McGrew" <[EMAIL PROTECTED]> wrote:
> I've mentioned Generator, and there are also
> several others that allow the user to define the
> problem in an Excel spreadsheet.  
>
>I don't mean to pick a quarrel, but you sure seem
>to want one.

        I've got no interest in quarreling, and I assume that you and
I have basically the same interest with respect to GAs, which is to
help educate people about what GAs are and how to use them to solve
practical problems.  I'm offering that help to non-programmers in
particular.

...

>An "Excel" spreadsheet solution might be dandy for
>tinker toy problems, and I count some pretty
>damned impressive ones as "tinker toy", but even
>mine tend to use on the order of a hundred
>different approaches and tens of thousands of
>lines of code to solve one single instance of one
>easily described kind of problem that happens also
>to be incredibly difficult, and Microsoft has never
>made nor sold a product sufficient to withstand
>that kind of assault in its existence.  "Generator"
>is not the all-purpose answer to every GA query, so
>stop offering it as such, while you still have
>Net connectivity.

      Even though Excel is encumbered by the Windows operating system
and various other overheads, it does have a lot of useful features
such as the spreadsheet interface, graphing tools, etc. It's really,
really convenient to define a problem in Excel for solution by a GA,
without having to write a line of code.  In most of the problems I've
solved using a GA, setting up the problem takes considerably longer
than solving the problem, so it seems worthwhile to find ways minimize
the effort the user needs to exert to set up a problem.

      Is the problem instance you refer to above a TSP?  Aside from
the machine cycles wasted on Windows overhead, Generator working
through Excel has had no particular difficulties with large problems
including TSPs, optical design problems, etc, with thousands of
variables.  No doubt a custom-coded GA running under Linux can solve
the problem faster, but I'd wager that the elapsed time from starting
to define the problem to seeing the finished solution is shorter when
the user doesn't have to write code.

Steve