Re: [Nolug] Mimedefang Spamassassin Bayesian learning

From: -ray <ray_at_ops.selu.edu>
Date: Thu, 18 Dec 2003 10:37:57 -0600 (CST)
Message-ID: <Pine.LNX.4.44.0312181029330.13804-100000@romulus.csd.selu.edu>

On Thu, 18 Dec 2003, Dustin Puryear wrote:

> > I don't use Bayes with SA, after reading the docs I felt it was too much
> > effort to keep it accurate, and you should only use it if you're gonna
> spend
> > a lot of time keeping it accurate. So unless I was maintaining an
>
> That's not really true. You do have to initially train the bayes, but once
> trained you can leave it alone. Just be sure to either set the auto-learning
> values to both very high for spamminess and very low for hamminess (the
> defaults are pretty good as is), or disable auto-learning entirely.

You don't even have to train it, that's the point of auto-learning! The
Bayes db will train itself based on what Spamassassin considers spam
(score > 12) and ham (score < 0.1). But the before the Bayes rules will
kick in and start affecting the total score, you do have to have a good
amount of spam and ham in the Bayes db. The defaults are 200 ham and 200
spam messages. All those default values are configurable in the config
file.

ray

-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Ray DeJean  				       	 http://www.r-a-y.org
Systems Engineer                    Southeastern Louisiana University
IBM Certified Specialist  	      AIX Administration, AIX Support
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
___________________
Nolug mailing list
nolug@nolug.org
Received on 12/18/03

This archive was generated by hypermail 2.2.0 : 12/19/08 EST