Michael Moncur wrote: > > Right now there's test for spam phrase scores of 10, 20, 40, and 100. Why > > aren't there test for, say, 15, 25, and 30? If we're going to test for 40 > > and 100, shouldn't there also be a test for halfway in-between at > > 70? Is the > > current set of scores optimal in some way that I can't see?
Spam phrases are voodoo -- no doubt 10, 20, 40 and 100 is just part of the incantation. > Actually the current scores are about as non-optimal as they could possibly > get: > > score SPAM_PHRASES_020 0.0 > score SPAM_PHRASES_030 0.0 > score SPAM_PHRASES_100 0.0 > (all from 50_scores.cf in the latest CVS) > > Is the spam phrases system still broken? Or were these left turned off by > mistake? Updating spam phrases depends to a large extent on having access to good non-spam corpus. When I updated the scores in the latest release, I think I turned off phrase checks, but I don't recall why now. Could have been a late night, or might have been a problem. I'll turn them back on and see if I can remember why. > I believe the issues with really short messages getting high scores were fixed > and the system should be worth *something* now... Yes -- it should work OK now. The reason I turned it off is probably because the phrases needed regenerating, but I didn't have a good nonspam-corpus to generate from. C _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
