Re: DM: Minimal rule covering

[e]

At 03:38 PM 10/9/98 -0300, Jose Augusto Baranauskas wrote:
>Hi!
>
>I'd like to know about about minimal subset rule covering 
>algorithms. I
>mean, after genetaring lots of induced rules, I'd like to get a 
>minimal
>robust subset that covers the majority of instances. 


there's a chapter on this topic in my book

http://rrt.base.org/book.html
(seems to be down at the moment; try
 http://www.geocities.com/ResearchTriangle/2441/book.html )
with discussion of prior work.

i develop some criteria for what it may mean to be "minimal",
and some algorithms that can be proven to optimize
these criteria (with some caveats, naturally). 
some of these algos are implemented as part of the
theory restructuring module in Mobal

http://nathan.gmd.de/projects/ml/mobal/mobal.html
which you can download & use if you have a Sun running
Solaris (Chandler Caulkins <chandler@icmc.sc.usp.br>
has done this if you'd like to confer with someone
closer to home). Be warned, however, that Mobal is
a mature system, meaning it was developed from 1990 to
1996 and is not supported ..


you might also do a lit search on "redundancy" and "pruning".

several ML algos contain pruning or reduction steps
as post-processing, e.g. Quinlan's C4.5 & FOIL and Muggleton's Golem.


have fun!
eddi



 ______________________________________________________________

 Dr. E Sommer                                 +1 (868) 639-4213
 Senior Consultant       IntelliCorp                      -2452