200 of each. The choice of 200 is slightly arbitrary, but the basic gist is that SA can't effectively use bayes until it's seen a reasonable sample of both kinds of email.
In the extreme, if you tried to use bayes after you gave it 200 spams and 0 ham messages, everything would look like it had a 100% chance of being spam, since there would be no ham tokens at all.
Thanks. I guess false positives will be tagged SPAM and not be automatically deleted?
rewrite_header Subject [SPAM] report_safe 0 use_bayes 1 auto_learn 1
Regards, Norman Zhang
