The rules has more weigh than bayesian-learn ?

[Thiago Henrique Rodrigues <th...@digirati.com.br>]

Dear,

I am a novice in the use of SpamAssassin and I need your help. Who has
more weigh in the classification of a message, the rules or the
bayesian-learn ?

Best Regards,

--
[]'s
Thiago Henrique
Network Administration
Digirati Networks
K8 Networks
Hostnet Hosting

RE: The rules has more weigh than bayesian-learn ?

[Giampaolo Tomassoni <g....@libero.it>]

> -----Original Message-----
> From: Thiago Henrique Rodrigues [mailto:thiagoh@digirati.com.br]
> Sent: Thursday, June 19, 2008 11:11 PM
> To: users@spamassassin.apache.org
> Subject: Re: The rules has more weigh than bayesian-learn ?
> 
> 
> > Your question doesn't really make sense.  The results of the Bayes
> > examination
> > are rules based on the 0-100 spam probability.
> >
> > If I understand what you're asking though, the Bayes system results
> in
> > 1 rule
> > hit, whereas there are hundreds of other rules that can all hit, so
> > generally
> > rules would outweigh Bayes, unless you change the weighting (score)
> of
> > the
> > Bayes rule in relation to the other rules.
> >
> 
> Thanks for helping. I didn´t understand until now. All make sense.
> 
> I'm trying to use SpamAssassin in a structure as such: Postfix + Amavis
> + Clamav + SpamAssassin. Will I lose much considerably in the quality
> of
> my anti-spam if I not use the bayesian rule?

Bayes is a good piece of code, which fits fine in SA. Thereby I would suggest not to avoid using it.

Nevertheless, if you still prefer not to rely on it, you may lower a bit the spam tag and kill levels in Amavis such that pattern rules and network tests may suffice.

I find bayes quite useful also in avoiding FPs, not only in detecting spam...

Giampaolo


> 
> Best Regards,
> 
> --
> []'s
> Thiago Henrique
> Network Administration
> Digirati Networks
> K8 Networks
> Hostnet Hosting
> 
> 
> 

Re: The rules has more weigh than bayesian-learn ?

[Thiago Henrique Rodrigues <th...@digirati.com.br>]

> Your question doesn't really make sense.  The results of the Bayes
> examination
> are rules based on the 0-100 spam probability.
> 
> If I understand what you're asking though, the Bayes system results in
> 1 rule
> hit, whereas there are hundreds of other rules that can all hit, so
> generally
> rules would outweigh Bayes, unless you change the weighting (score) of
> the
> Bayes rule in relation to the other rules.
> 

Thanks for helping. I didn´t understand until now. All make sense.

I'm trying to use SpamAssassin in a structure as such: Postfix + Amavis
+ Clamav + SpamAssassin. Will I lose much considerably in the quality of
my anti-spam if I not use the bayesian rule?

Best Regards,

--
[]'s
Thiago Henrique
Network Administration
Digirati Networks
K8 Networks
Hostnet Hosting

Re: The rules has more weigh than bayesian-learn ?

[Theo Van Dinter <fe...@apache.org>]

On Tue, Jun 17, 2008 at 04:32:00PM -0300, Thiago Henrique Rodrigues wrote:
> I am a novice in the use of SpamAssassin and I need your help. Who has
> more weigh in the classification of a message, the rules or the
> bayesian-learn ?

Your question doesn't really make sense.  The results of the Bayes examination
are rules based on the 0-100 spam probability.

If I understand what you're asking though, the Bayes system results in 1 rule
hit, whereas there are hundreds of other rules that can all hit, so generally
rules would outweigh Bayes, unless you change the weighting (score) of the
Bayes rule in relation to the other rules.

-- 
Randomly Selected Tagline:
"I am NOT a computer geek! ... I just spend too much time in front of the 
 computer."                  - Theo

Re: The rules has more weigh than bayesian-learn ?

[Kathryn Kleinschafer <ka...@rheel.co.nz>]

Thiago Henrique Rodrigues wrote:
> Dear,
>
> I am a novice in the use of SpamAssassin and I need your help. Who has
> more weigh in the classification of a message, the rules or the
> bayesian-learn ?
>
> Best Regards,
>
> --
> []'s
> Thiago Henrique
> Network Administration
> Digirati Networks
> K8 Networks
> Hostnet Hosting
>
>   
It depends on the scores you give each thing. I.E bayesian confidence 
90-100% you might give 6 points to and a single rule onlly 1 point. But 
if the bayesian confidence was only 5% you may give it 0 points.

Kate