Hi, >> If so, can I pull the messages out, unlearn them, then re-learn them >> as spam instead? > > I don't think this makes sense. You don't know which messages biased your > db in the wrong way. It's better to start over.
Bayes has such a substantial role in the messages that get tagged, I couldn't just dump it, for the time it would take to rebuild the db, it's likely that too many messages would get through, no? I meant that if I "--forget" a message that isn't in the db, then it's no harm, but if I do it on a message that has tokens in the db, then re-learn them the right way, wouldn't that be correcting the db? I could easily collect 1000 ham and 1000 spam and use that as a basis for rebuilding the db. Would that be sufficient? > And you may want to check if you auto-learn many spam messages as ham. How can I do that? Thanks, Alex
