From: JamesDR [mailto:[EMAIL PROTECTED] > > Mike Pepe wrote: > > I've been feeding messages from my inbox into a folder that SA > > reads as ham for quite some time now. > > > > Suddenly it occurs to me that this may be a bad idea, and I should > > only have SA learn messages as ham that it believes is spam. > > > > This strikes me as being as bad as forcing SA to re-learn spam as > > spam again. > > > > Am I correct in this assumption, or is re-learning good email as > > ham safe? > > > > I'm hoping that I can finally put an end to these new and > > especially annoying "timepiece" emails that sneak through. > > If autolearn is enabled, and it is doing its job, then it should be > ignoring the mails. However, if you don't have autolearn enabled, > then training as ham/spam on correctly marked mails isn't a bad > thing (IMO.) Keeps the DB fresh with the current mail stream. I > train everything that was classified correctly/incorrectly and has > worked well for years. Others may have other opinions. It appears, > at least to me, the fresher the bayes db, the better results out of > bayes. As usual, YMMV
As James said, it is never a bad thing to train Bayes. At best, it learns something, at worst, it ignores the training attempt. Bayes will remember the messages it has already learned. This allows it to avoid learning the same message twice. Personally, I will train Bayes on every message until I get the database to a useable size. Then, I switch to training only on mistakes. This has worked well for me. Bowie
