Tom, IANABE (I am not a bayesian expert) but with a naive understanding of the algorithm, I think I can see value in learning an email that's already scored at 1.000.
There were obviously a lot of spammy tokens in that email which bumped the score to 1. However, even an email that scores 1 can have tokens that are neither spammy nor hammy, or tokens not yet seen. Learning this email would not only reinforce the super-spammy tokens (questionable value), but also help weight the somewhat neutral tokens towards spammy and begin to weight the never seen tokens. Disclaimer: I'm probably wrong. Chris Tom Meunier said: > If it's already 100% sure that it's spam, how is it helpful to train it > that it's spam? It's not like it's going to be 110% sure that it's > spam. It's already trained! > > Not trying to be a wise-ass, I've just seen this question come up fairly > often, and can't wrap my head around it. > > -tom > >> -----Original Message----- >> From: [EMAIL PROTECTED] >> [mailto:[EMAIL PROTECTED] On >> Behalf Of Chris Barnes >> Sent: Wednesday, November 05, 2003 2:09 PM >> To: [EMAIL PROTECTED] >> Subject: [SAtalk] a new rule >> >> How hard would it be to create a new rule for BAYES scoring >> that IS used by autolearn? >> >> Specifically, when I see this: >> * BAYES_99 BODY: Bayesian spam probability is 99 to 100% >> * [score: 1.0000] >> >> in the header, I'm fairly comfortable with having it autolearnt and >> letting my .procmailrc script send it straight to /dev/null. >> However, >> since BAYES scores are not used in deciding whether or not >> autolearn is used, this rarely happens. >> >> So a new rule of BAYES_100 (for scores that are 100%), that >> is used would be helpful. How? >> > > > ------------------------------------------------------- > This SF.net email is sponsored by: SF.net Giveback Program. > Does SourceForge.net help you be more productive? Does it > help you create better code? SHARE THE LOVE, and help us help > YOU! Click Here: http://sourceforge.net/donate/ > _______________________________________________ > Spamassassin-talk mailing list > [EMAIL PROTECTED] > https://lists.sourceforge.net/lists/listinfo/spamassassin-talk > > ------------------------------------------------------- This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/ _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
