John,
I also experienced the exact same thing. The default setup with no bayes
training really let most of my spam through our filters.
I chose to let the auto training of bayes to work and it did well since I get
to see at least 10,000 spams a day. Along with adding additional cf's I have
moved my threshold back towards 5.0. I'm still sitting at 3.5, but will
eventually probably move it to 5.0.
For the newbie thomas, I also watched what spam was getting under my threshold
and then adjusted I'd say about 2 dozen of the individual scores to help bring
the low scoring spam up nearer to 5.0. You'll have to watch which ones you up
though ie: I brought HTML_MESSAGE up too high and started getting a lot of
false positives.
Good Luck
Scott
-----Original Message-----
From: John Fleming [mailto:[EMAIL PROTECTED]
Sent: Tue 3/2/2004 6:17 AM
To: [EMAIL PROTECTED]; [EMAIL PROTECTED]
Cc:
Subject: Re: NEWBIE: Not extremely happy with SA filtering so far
----- Original Message -----
From: "Thomas Muller" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Tuesday, March 02, 2004 6:10 AM
Subject: NEWBIE: Not extremely happy with SA filtering so far
> SpamAssassins,
>
> I've been trying out CommuniGate in conjunction with SpamAssassin for
a
> while, and I must say my old client-side filter (Spammunition) caught
a
> lot more spams than the default setup of SpamAssassin are doing now. I
> have the overall score threshold set to 5.0, but a lot of spam has
only
> a score of 2-3 anyway. I probably have to tweak it a lot more in order
> to catch the subset of spam aimed to my addresses.
>
> In particular I need some more rules that can take all the viagra
> variants (like Generi*c V-iagra Half Price Guaran-teed etc).
>
> Forgive me being a total novice, is it a central storage where rules
can
> be downloaded and deployed (probably need a how-to description as
well)?
>
> Any input/comments would be much appreciated.
>
I'm new too - Initially I settled on a threshhold of 2.4, and this got
most
things with a few false positives. Then I trained my Bayes with 400
spams
and 200-some hams and added most of the rules from
http://www.merchantsoverseas.com/wwwroot/gorilla/sa_rules.htm Now my
threshhold is back up to 5.0 and it's almost perfect for me! Good luck
-
John
