Default tweaks?

[Chris Withers <li...@simplistix.co.uk>]

Hi again,

With my nice shiny new SpamAssassin install, I'm noticing that it's not doing 
very well, still getting about 30% of the spam coming through the net :-S

What are the default tweaks people make (apart from sa-learn, which I'll be 
using once I've accumulated a couple thousand more spams) to the rules and 
weightings to get it catching more?

cheers,

Chris - SA Conf Noob

-- 
Simplistix - Content Management, Zope & Python Consulting
            - http://www.simplistix.co.uk

Re: Default tweaks?

[Bob George <ma...@ttlexceeded.com>]

Chris Withers <li...@simplistix.co.uk> wrote:
> Bob George wrote:
>
>> I definitely recommend those on the rules emporium and the
>> rules_du_jour script.
>
> Where can I find these?

http://www.exit0.us/index.php/RulesDuJour  &
http://www.merchantsoverseas.com/wwwroot/gorilla/sa_rules.htm

(and of course, google with appropriate keywords to find more).

- Bob

Re: Default tweaks?

[Chris Withers <li...@simplistix.co.uk>]

Bob George wrote:

> I definitely recommend those on the rules emporium and the rules_du_jour
> script. 

Where can I find these?

cheers,

Chris

-- 
Simplistix - Content Management, Zope & Python Consulting
            - http://www.simplistix.co.uk

Re: Default tweaks?

[Bob George <ma...@ttlexceeded.com>]

Keith C. Ivey <kc...@cpcug.org> wrote:
> Bob George <ma...@ttlexceeded.com> wrote:
> [...]
> Actually the default boundary for the ham autolearning is 0.1,
> not -2, so it can be dangerous.  Of course if it were -2 you
> would never autolearn much ham, unless you added a bunch of
> negative-scoring custom rules.

Ah, thanks for the correction. I did NOT recall correctly. The defaults have
worked well for me in any case.

Are you saying 0.1 (default) IS dangerous though?

- Bob

Re: Default tweaks?

["Keith C. Ivey" <kc...@cpcug.org>]

Bob George <ma...@ttlexceeded.com> wrote:

> I fed sa-learn several large inboxes of ham (skimmed for borderline
> cases) and started collecting spam-flagged messages WITHOUT any
> auto-learning enabled initially. The defaults are (IIRC) +12 for
> auto-learning spam, and -2 for ham, so they're pretty safe even if left
> on initially.

Actually the default boundary for the ham autolearning is 0.1, 
not -2, so it can be dangerous.  Of course if it were -2 you 
would never autolearn much ham, unless you added a bunch of 
negative-scoring custom rules.

-- 
Keith C. Ivey <kc...@cpcug.org>
Washington, DC

Re: Default tweaks?

[Bob George <ma...@ttlexceeded.com>]

Chris Withers <li...@simplistix.co.uk> wrote:
> [...]
> With my nice shiny new SpamAssassin install, I'm noticing that
> it's not doing very well, still getting about 30% of the spam
> coming through the net :-S

Through the ENTIRE net? Sheesh! :)

Seems like it sometimes, I'm sure.

Are those that do get through at least "close" in terms of scoring?

> What are the default tweaks people make (apart from sa-learn,
> which I'll be using once I've accumulated a couple thousand
> more spams) to the rules and weightings to get it catching
> more?

I definitely recommend those on the rules emporium and the rules_du_jour
script. Those have helped with getting bayes up to speed for me. They'll help
kick the borderline cases up just enough to be properly marked. You can also up
the scores for those that are hitting, but do so carefully.

I actually started to see positive results with a FEW HUNDRED spam/ham, so long
as I was diligent about training. Constantly feed back any mis-classified
messages, and it should start working. I've found it does a very good job on
allowing "spammy" mailings that I DO want, while stopping the rest.

DO make a point of reviewing what you feed it initially for best results. I fed
sa-learn several large inboxes of ham (skimmed for borderline cases) and
started collecting spam-flagged messages WITHOUT any auto-learning enabled
initially. The defaults are (IIRC) +12 for auto-learning spam, and -2 for ham,
so they're pretty safe even if left on initially. But feed it spams < 12 as
well!

Good luck with it.

- Bob