d:dev@spamassassin.apache.org 23 July 2005 • 9:08PM +0800

notes from brief chat with Andrew Ng
by Daniel Quinlan

Reply to author Reply to group

Andrew suggests we look into LR and, more specifically, l1 LR, which is  
a technique that does something like:  
 
  x1 + x2 + x1x2 + ...  
 
The l1 modification of LR generates useful meta-like rules that should  
not be too overoptimized for SA.  
 
So, we should try both.  :-)  
 
Daniel  
 
--  
Daniel Quinlan  

openSubscriber is not affiliated with the authors of this message nor responsible for its content.

Copyright © openSubscriber.com 2004-2006
All rights reserved.