Test and Keep spam

[Kenneth Porter <sh...@sewingwitch.com>]

Been getting a bunch of these lately, and they're falling on either side of 
the 5.0 margin. Two that came in under 5.0 today have unusual 
characteristics: The Bayes score on one is 60% and scores higher than one 
with an 80% Bayes score. You can see my current "uncaught" corpus here:

<http://home.sewingwitch.com:8000/Stuff/UncaughtSpam.mbox>

Re: Test and Keep spam

[alan premselaar <al...@12inch.com>]

Kenneth Porter wrote:
> Been getting a bunch of these lately, and they're falling on either side 
> of the 5.0 margin. Two that came in under 5.0 today have unusual 
> characteristics: The Bayes score on one is 60% and scores higher than 
> one with an 80% Bayes score. You can see my current "uncaught" corpus here:
> 
> <http://home.sewingwitch.com:8000/Stuff/UncaughtSpam.mbox>

Kenneth,

   I've noticed with my corpus that BAYES_95 and BAYES_99 score less 
than say BAYES_80 ... which has been a little discouraging for me since 
most of the mail i'm filtering is japanese and other test don't hit 
often so I have to rely heavily on my (manually trained) Bayes database...

having items that hit BAYES_99 only scoring 1.8 and change compared to 
the 2 and change that BAYES_80 scores has been a little frustrating.

I'm tempted to change the scores for BAYES_95 and BAYES_99, but i'm 
concerned about what other effects that might have ...

not sure if this information will be helpful or not, but thought i'd 
share anyways.

alan

p.s. I'm using SA 3.01 with MIMEDefang 2.49 on this machine. no 3rd 
party rulesets installed.