[Mimedefang] Sitewide Bayes

[cam]

> I disagree.  Unless your users are technically savvy, the average customer
> will never use Bayesian rules.
> 
> Therefore, I think it comes down to: is Bayesian site-wide better for a
> server where the users would never individually seed the bayesian filters?
> 
> KAM
> ----- Original Message ----- 
> From: "Lucas Albers" <admin at cs.montana.edu>
> 
> > Imo
> > It is not prudent to run a single site-wide bayes for your mail server.
> > As David said awhile ago, every one considers different mail as spam.
> > So having a account to forward mail to as spam for bayesian learning,
> > shoule have a fairly high false-positive rate.



along the lines that KAM mentioned, what about users that NEVER log into a 
shell?  ie, (l)users on windoze?  they wouldnt' be able to set it up 99% 
of the time due to a lack of desire to learn.

i personally use the bayesian site-wide solution myself, but also make use 
of all_spam_to config options for a few users who bitched.  ( hmmm about 3 
people out of 2000  if i remember right. )

cam proctor
bcp1 at craqskeep.com
bcp1 at kopower.com