[EMAIL PROTECTED] said: :-What do others on the list think of this? I think the "pipe" action as you describe would work.
There is another aspect of Bayesian filters: construction of the spam and non-spam corpus. TMDA can help here too. Messages that make it through TMDA are presumably a good set to collect as a non-spam corpus. Messages that are dropped or are never confirmed are presumably spam and are a good spam corpus. You can use the "program" delivery instruction to collect a copy of the good (non-spam) message for your non-spam collection, although it would be nice if you could somehow achieve that when using the mbox or maildir instructions. I don't know how you collect messages from "drop" actions. Can you? Also you need to collect the messages that never get confirmed because they are presumably spam. I would use tmda-pending for that, but I don't know how to collect messages that I --delete using tmda-pending. Thanks! _____________________________________________ tmda-users mailing list ([EMAIL PROTECTED]) http://tmda.net/lists/listinfo/tmda-users
