>-----Original Message----- >From: Jeff Chan [mailto:[EMAIL PROTECTED] >Sent: Friday, February 18, 2005 12:30 AM >To: SpamAssassin Users >Subject: Re: Less spam blocked with 3.02 - AWL-related? > > >On Thursday, February 17, 2005, 6:45:58 AM, Juergen Georgi wrote: >> Score set 3 - the fourth score - applies when Bayes is enabled and >> network tests are enabled. > >> In addition, non-zero score set 3 values in 3.0.2 are much lower >> than those in 2.64, e.g. > >> 2.64: score SORTED_RECIPS 4.299 4.300 2.696 2.699 >> 3.0.2: score SORTED_RECIPS 0.879 1.155 1.759 0.887 > >> 2.64: score DEAR_FRIEND 1.888 1.065 2.397 1.846 >> 3.0.2: score DEAR_FRIEND 0.542 0.766 1.288 0.070 > >> This is probably intented > >Yes, it's intentional. > >> - but I just don't understand why so many >> syntactical checks are completely disabled or strongly damped when >> using Bayes and network tests together. > >My understanding is that the scores are optimized automatically >using a perceptron. Probably it's mentioned in the SA wiki or faq.
Yeah that perceptron still has me confused. Although I haven't put enough time into it yet. It *think* it simply tries to push the average of scores towards 5.0. I think. Maybe? Eh... --Chris (It is time Mr Freeman...)
