On Saturday 30 August 2003 11:48 pm, Steve Lamb wrote:
>On Sat, 30 Aug 2003 23:40:13 -0400
>
>Tom Allison <[EMAIL PROTECTED]> wrote:
>> It may be turned on in the config files, but I am guessing that the code
>> is skipping the bayesian score contribution until the mail count gets to
>> 200 on each side (ham/spam).
>
> Right.
>
>> I just grabbed a lot of email I had already and fed it into the sa-learn.
>> I think I have enough now that it is working.
>
> You can tell by looking at the headers and seeing if BAYES_xx shows up.
>The xx is the approx. range that the Bayesian filter places the particular
>piece of mail. For example here's the score from the message of yours I am
>responding to:
>
>X-Spam-Status: No, hits=-3.6 required=5.0
> tests=BAYES_10,NO_REAL_NAME
> version=2.55
>
> So the Bayesian filter (classifier?) thinks it is 10-??% (forget the
> upper range) likely to be spam. Ah, here it is. From 23_bayes.cf...
>
>body BAYES_10 eval:check_bayes('0.10', '0.20')
>
> ...10 to 20% which gives it a score of...
>
>score BAYES_10 0 0 -5.300 -4.701
>
> ...-4.701 based on my setup. IIRC first score is if no network checks
> are enabled, second score is if network checks are enabled. Well, let's
> see. NO_REAL_NAME nets the message...
>
>score NO_REAL_NAME 0.993 0.820 1.137 1.149
>
> ...1.149. -4.7 + 1.1 = -3.6
>
>> I'm not sure, I just kind of fiddled with it a few times in the early
>> hours and got it working.
>
> Yeah, it just takes a little bit to kick in. Once it does the
> difference is dramatic if you track the scores. Average ham for me is
> around -3 and average spam is closer to 12 to 15. Affords me a lot of
> latitude when configuring sa-exim to reject things at SMTP.
I'm learning a lot here. This is a valuable thread, thanks!
Jeff
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
