Am 16.02.2015 um 21:10 schrieb Amir Caspi:
On Feb 16, 2015, at 1:01 PM, RW <rwmailli...@googlemail.com> wrote:IIWY I'd look into rescoring the BAYES_* rules.I was already rescoring them as BAYES_99 = 4.0, BAYES_999 = 0.5 ... so a total score of 4.5 if both rules hit. These FNs typically get scores of 4.6, so the other rules are simply not good enough. Since I've basically never seen an FP caused by BAYES_99/999 unless I accidentally mistrained it (e.g., sent a newsletter through it by accident), this morning I increased the scoring for BAYES_99 to 4.4, leaving BAYES_999 as 0.5... so now hitting both of them will yield a score of 4.9, and the remaining 0.1 from the other miscellaneous tests should push it over to 5.0. (I didn't want to make BAYES_999 a complete poison pill... though it's darn close.) I guess we'll see if that's enough, though now I'll have to monitor more carefully for FPs for a while...
we use the scores below by a milter reject of 8.0 i can't rememeber a single FP over 6 months caused by bayes most likely because the large ham-corpus combined by a lot of DNSWL score BAYES_00 -3.5 score BAYES_05 -1.5 score BAYES_20 -0.5 score BAYES_40 -0.1 score BAYES_50 2.0 score BAYES_60 3.5 score BAYES_80 5.0 score BAYES_95 6.5 score BAYES_99 7.5 score BAYES_999 0.4
signature.asc
Description: OpenPGP digital signature
