HTML Spam and other spam

[Mark Nienberg]

I agree that Bayes made a big difference for us, and I too don't bother with the sa-
learn stuff.  Initially, you may want to add a line in your spam.assassin.prefs.conf file
for:

auto_learn_threshold_nonspam 0.3

The default value of -2.0 (yes, minus 2.0) for auto learning nonspam is too low to
generate enough ham, because not many rules have negative scores.  After you get
enough spam and ham, bayes starts working, and eventually you can revert to the
default or some other value. Note that bayes does yield negative scores for
messages that have a low probability of being spam, but this won't happen initially
because bayes doesn't kick in until it has enough spam to evaluate.

DCC has also helped quite a bit.  I'm so pleased with it that I am working now on
installing pyzor, which seems a bit harder to get going, but I hope for success soon.

By the way, I noticed that with the 2.60 release of SA, there are default values for
Spamcop RBL in the scoring file.  They are smaller than the 4.0 that is listed in
spam.assassin.prefs.conf.  Presumably, you should remove the line in
spam.assassin.prefs.conf if you want to achieve the scoring accuracy that SA claims.

Mark

On 31 Oct 2003 at 17:33, Peter Bonivart wrote:

> I use Bayes and DCC (instead of Razor) and they do an amazing job. They
> both trigger on almost all spam and together with the Spamcop RBL which
> is pretty aggressive they add 12 points. Bye bye spam.
>
> There's a lot of talk about the trouble learning Bayes but some that
> don't run Bayes might have missed that it learns itself anyway. Messages
> with very low score are considered ham and messages with very high score
> are considered spam. I don't use sa-learn and I still achieve amazing
> results. As I said it tags almost every spam message and it's 99% sure
> it's spam so that's a cool 5.4 points. It's a shame if people don't use
> Bayes because they think it's high maintenance.
>
> DCC is also really good and is easy to setup. Check the INSTALL file
> that comes with SA (or on the web). The instructions are on four lines.
>
> /Peter Bonivart
>
> --Unix lovers do it in the Sun
>
> Sun Fire V210, Solaris 9, MailScanner 4.23-11, SpamAssassin 2.60 + DCC
> 1.2.9, ClamAV 20030829
>
> Juan C. Quesada wrote:
> > Im runing the latest Mailscanner with spamassassin
> > 2.60. I have not configured Bayes or razor, nor do I
> > know how to do sa-learn.