Which I'm sure is no surprise to most. Just wanted to mention that I'd been trying to clean up a polluted ham corpus by using 'forget', but I continued to get poor results. Long story short, I cleaned up my ham, and removed both ham, and spam that had spambouncer headers - which I assume sa-learn did not remove? (not that I would expect it to), deleted bayes database, and retrained.
Bayes is now much more accurate - even before turning on bayes, a few spam had begun to trickle through. I got zero spam today, and even caught, out of around 400 spam, 2 that would have otherwise slipped in. The reason I mention it - aside from being pleased - is to point out that it appears the problem was either the old spambouncer headers, or forgetting, wasn't (the latter being what I started out suspecting, until I discovered the spambouncer headers). Another imperfection in my experience analysis is that I suppose the database could have been corrupted, as I'd gotten knocked off line several times while running sa-learn. In any event, if trouble training, might want to consider starting from scratch. Bryan ------------------------------------------------------- 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
