sasa wrote: > Hi (I use spamassassin-3.0.4), I have tried this on my account where > I send all spam mail then have arrived on other my accounts: > > #sa-learn --spam --mbox /var/spool/mail/spam-report > Learned from 18 message(s) (19 message(s) examined). > > ..but however also after this operation the same spam mail aren't > stopped and delivered to my account!! > where I mistake ?? > thanks. >
Hi Salvatore, sa-learn only feeds the emails to the Bayes training engine. This doesn't guarantee that the same email will be marked as spam in the future. There might have been other copies of the same or similar email that were incorrectly learned as ham, tilting the score towards ham. Also, Bayes score alone will not mark an email as spam in most cases. I wonder if your system of forwarding spams to that mbox might be changing the headers; many mail clients do that, most notoriously Outlook. That would greatly decrease the effectiveness of sa-learn, since Bayes considers both headers and message body. Good luck Pierre Thomson BIC
