JD Smith wrote: > That is what I was beginning to suspect. Is there a way to untrain the > emails I ran through it?
Feed them to sa-learn --forget. >It was a pretty large selection.. A few > thousand of both spam and ham. > > I turned on auto-learning so it should start to pick things up on it's > own without needing me to train it, no? True, but it's generally best to start off with at least a small sample of hand-picked emails for training. Left entirely to it's own devices, bayes learning can sometimes go awry. A little hand training helps keep it from mis-learning because of the "no contradictions" rule. (Never autolearn as spam any message that existing training says is strongly ham, and vice versa). > > Training on a per user basis could be difficult as this is a gateway > scanning and feeding mail on to around ten or so domains some of which > are rather large such as the Caddo Parish School Board that puts through > around 10,000 mails per day. If you're using MailScanner, there's no option for per-user bayes.
