Hi Richard, Thank you for a thoughtful reply. I'm ok with collecting spam. The only thing is that I read in the docs somewhere that this collected spam has to be pretty recent. Or are they saying that both ham and spam collected should be from the same time period. If so, do you know what the maximum time discrepancy could be? Should I worry about it at all?
Thank you! Sergei On Tue, Apr 11, 2006 at 09:18:33AM -0400, BMWrider wrote: > Sergei, > > Collect the SPAM that is getting through now in a folder. From a root > account run the following: > > sa-learn --spam /Path to folder containing SPAM > > You can train SA with any number of SPAM messages. The overall > accuracy of bayes will improve after > it has a large corpus, but you can rather quickly stop specific SPAM > messages by training manually. > Manual training is important because you are training with the SPAM > hitting your server. > > The whole thing is a cat and mouse game between SA users and > spammers. You stop them today > and they will come up with something new tomorrow. Then you will have > to manually train again and > the cycle continues. > > You can do the reverse if SA tags ham as SPAM. Feed the good email > back to SA with > > sa-learn --ham /Path to folder containing HAM > > Richard > > > > > > > > > "The advantage of a bad memory is that one enjoys several times the > same good things for the first time." > ....Friedrich Nietzsche > > > On Apr 10, 2006, at 10:00 PM, Sergei wrote: > > >I tried sa-learn, but don't you need a sizable spam collection for > >it to > >work? The docs say that you need to collect about a thousand of ham > >and > >spam messages before the training starts to work. That sounds like a > >pain in the neck. Or am I missing something? > > > >I ran sa-learn on this one message, than ran SA with the -D switch and > >it said that it was ignoring the Bayes database because there was only > >one message. > > > >How is it working for you? Or did you do the initial training? > > > >On Mon, 2006-04-10 at 21:49 -0400, BMWrider wrote: > >>All, > >> > >>When I upgraded to 3.1.1 the Online Pharmaceutical SPAM also came > >>through. It didn't take much effort to > >>run sa-learn --spam on a bunch of them to shut them out. The spammers > >>are trying some new tricks right > >>now which will get through a fresh upgrade but again manual training > >>will stop them quickly. > >> > >>Richard > >> > >> > >> > >>"The advantage of a bad memory is that one enjoys several times the > >>same good things for the first time." > >>....Friedrich Nietzsche > >> > >> > >>On Apr 10, 2006, at 8:59 PM, Sergei Gerasenko wrote: > >> > >>>Thanks for such a quick reply. So upgrading would really be > >>>helpful in > >>>terms of performance if nothing else. Ok, I'll give it a thought. > >>>Maybe > >>>I'll find a Debian package with the latest version. Should be > >>>possible. > >>> > >>>I installed SpamAssassin today for the first time and "The Ultimate > >>>Online Pharmaceutical" (seems like a LOT of people get this one in > >>>particular) came through undetected. I had to add a manual rule to > >>>take > >>>care of it. Could that have happened because I have an older > >>>version of > >>>SA? If so, any options besides upgrading? > >>> > >>>Thanks! > >>> > >>>On Mon, Apr 10, 2006 at 08:40:03PM -0400, Matt Kettler wrote: > >>>>Sergei Gerasenko wrote: > >>>>>Hello everybody, > >>>>> > >>>>>Got a potentially previously answered question. I have > >>>>>spamassassin > >>>>>3.0.2-3, which is the current release with Debian. I wouldn't > >>>>>like to > >>>>>deviate from the official package and so I'm wondering if it's > >>>>>absolutely necessary to upgrade. I diffed the rules, they seem to > >>>>>be the > >>>>>same. > >>>>> > >>>>>That's actually why I'm looking into this. I'll need to update > >>>>>the rules > >>>>>periodically and sa-update is not in 3.0.2. Is there a repository > >>>>>of the > >>>>>standard rules somewhere? I couldn't find it no matter how hard I > >>>>>looked. > >>>>There are no standard rule updates that will work with the SA 3.0.x > >>>>codebase. > >>>> > >>>>The whole idea behind SA 0.1 through 3.0.5 was that if you > >>>>needed new > >>>>rules, you upgraded your SA version. Rule updates were previously > >>>>very > >>>>slow, due to the expensive mass-check process. New releases of SA > >>>>code > >>>>came out much faster than new rules, thus there was no point in > >>>>separating the two. (rule updates were typically only made once or > >>>>twice > >>>>for a given major.minor release of SA. ie: 2.60 and 2.64 had rule > >>>>updates, 2.61-63 did not.) > >>>> > >>>>With 3.1.1 and higher, the SA devs are trying out an approach of > >>>>adding > >>>>on rules and making updates to an already released version. > >>>>However, > >>>>this is a completely new concept, and thus only supported on the > >>>>completely new version. > >>>> > >> > > >
