> > > Chris Santerre wrote: >>>>The 1:1 ratio is a mistake based on a wrong interpretation of >>> >>>the bayes=20 >>> >>>>theorem. I have a ham : spam ratio of 1 : 40. >>> >> Also: >> "I thing bayes is a very good addition to individual rules. And when >> it's >> trained propper it works fine. " >> >> I agree on a personal scale it works wonders if you *continue* to feed >> it a >> proper diet. But when you get to a more general server side solution, I >> don't think the results are worth the effort, when one can write a >> simple >> rule faster then training. >> >> IMHO if it only works on a personal level, or a very small company >> level, >> its not a good solution. (Mainly because the starfish out there can't >> even >> work their microwaves, nevermind feed bayes!) >> >> --Chris > > Works great here at the ISP level, about 40K users. I let it auto learn > for awhile (a few weeks) then turn off the auto-learning and feed it > myself. > > Regards, > > Rick >
While i realize that my environment is not the norm (we're a hospital), we have had great sucess with bayes and autolearn. I initially trained bayes with a set of about 50/50 spam/ham and that was about it. Its autolearned the rest itself for almost a year now. We only get about 1k messages/day so i dont have a whole lot of volume to work with. Our spam (what little there is) is almost always tagged bayes_99, and the ham almost always bayes_0. I run quite a bit of net tests and i have tweaked the autolearn threshold slightly (i've lowered the ham to a negative number) but other than that, its a stock 2.64 install. I just dont see any need to upgrade or perform any other maintenance tasks...it just works. -Jim
