train spamassassin via shell by storing minimum your spam and maybe also your ham in extra mailboxes and read them on PMG via sa-learn. Ham is fine, but not highly necessary as spamassassin tries to auto-learn ham and it does the job really good. But spamassassin is very conservative in learning spam, so you should push spam to spamassassin by yourself. You need BTW a minimum of 200 spam and 200 ham to have spamassassins bayes filter working.
