On 08/11/03 05:12 PM, Hannu Liljemark sat at the `puter and typed: > On Mon, Aug 11, 2003 at 03:21:00PM +0300, Louis LeBlanc wrote: > > > > Most seem to score in BAYES_30 range in your current > > > environment, but with some training I'm willing to bet you'll see > > > BAYES_90 for the same mails. > > > > Always. Every day, I archive the spam folder of all the messages > > received the previous day. > > Okay, sounds good. I just thought it was weird to see the spams > matching BAYES_30 with your bayes db when here (with whatever tokens > the db holds currently, without sa-learn'ing your samples) the mails > hit BAYES_90.
You know, I am regularly pruning the bayes db, and I haven't gotten any of these particular messages until just recently. I suppose this particular tagset may have been cycled out of my db. It will probably pick back up once bayes has had a chance to relearn them. > > I use Razor and whatever else SA checks by default. Though I often > > see messages that were flagged as probably spam by one DSBL or > > another, but still fall under the 5.0 mark. > > Maybe the proxies were added to the RBLs very recently and weren't there > the moment you got the mails. And same with Razor; perhaps the checksums > for those mails weren't listed yet. Good point. In the past, I have seen messages score higher when rechecking them for this exact reason. Of course, there's no way to tell SA to wait and check again in an hour to see if someone else reported it (and who would want to do that anyway), so no sense worrying about it :) > Anyways with 2.55 the mails score high (I have added around +1 to > bunch of RBL tests, Razor and BAYES_90 scoring, though). I am running 2.55, but for the most part, I am using the defaults (only one whitelisted address, and one blacklisted address in my user_prefs). Overall, I've been very impressed with it's accuracy. Until these messages started coming about a week and a half ago, I was effectively getting about 100% accuracy, meaning one or two FNs per 1000 messages and only one FP since about a week after 2.55 was installed (that's when the bayes scores really started kicking in). Thanks for the feedback. Lou -- Louis LeBlanc [EMAIL PROTECTED] Fully Funded Hobbyist, KeySlapper Extrordinaire :) http://www.keyslapper.org ԿԬ Knebel's Law: It is now proved beyond doubt that smoking is one of the leading causes of statistics. ------------------------------------------------------- This SF.Net email sponsored by: Free pre-built ASP.NET sites including Data Reports, E-commerce, Portals, and Forums are available now. Download today and enter to win an XBOX or Visual Studio .NET. http://aspnet.click-url.com/go/psa00100003ave/direct;at.aspnet_072303_01/01 _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
