Re: fuzzy classification program?

Kamran Karimi <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...
> Hello,
> 
> I am searching for a freely available program for generating
> classification rules using fuzzy and/or genetic algorithms. The input
> could be a series of records like:
> 
> a,    b,           c
> 1,  blue,        true
> 0,  white,      true
> 3,  red,         false
> 
> where "a" and "b" are condition attributes and "c" is the decision
> attribute. The output could be a series of rules that predict "c" based
> on the values of "a" and "b."
> 
> Also, does anybody know of any temporal dataset with a format similar to
> the above? I mean a sequence of observations of a number of variables,
> preferably at regular intervals, over a certain period of time.

At ftp://users.aol.com/FlopsDemo/, you will find FlopsDemo.exe, a
doownloadable demo version of our expert system shell FLOPS that
contains a number of classification programs. Of special interest is
the program iris.par and its associated data base iris.dat, the famous
Fisher Iris classification data base.

When you download FlopsDemo.exe and run it, it will create a folder
FlopsDemoExport with two major programs: TFlops.exe, the integrated
program development environment, and Flops.exe, the run-time module
and inference engine. You must run TFlops.exe, load a sample program
from the Examples folder or write your own, and then run this program
from TFlops.

For more information there is a 100 page manual downloadable from
http://users.aol.com/wsiler/.

This will give you the tools for constructing classsification
programs. However, you will have to construct the classification rules
yourself. For any non-trivial classification problem, the idea that
you can simply run a training set into a program that will produce
your expert system classification program is probably an illusion,
though for simple problems this approach might work with a neural net.
You will have to think, even if thinking has gone out of style. The
progam examples should help you get started. The classification rules
in the Iris problem are very simple indeed. However, a rule-based
fuzzy expert system is a tool for emulating human reasoning with the
ability to del with uncertain or vague data as well as ambiguities and
contradictions, and reasoning tends to be more complicated than simple
"If (data) then (final conclusion)" rules. Consider the example
program Schizo.par; it would be virtually impossible to construct a
training set here that would cover all possibilities.

Please let me know how you do with these programs. Tech support is
available by EMail from me at no charge.

William Siler