On Sun, Mar 30, 2003 at 01:06:42PM -0500, Ross Vandegrift wrote: > Now, almost 100% of that ham is is autolearned, so I should have a > similarly strong weighting toward ham as Chris, right? > > This makes sense - I'm subscribed to a lot of high traffic mailing > lists, so I really do get a lot of legit mail. On the other hand, I'm > pretty good about storing a learning all my spam, so it seems it's a > problem.
Having a lot of heavy trafic mailing lists I had the same problem until I learned my last 6 months of spams and changed auto_learn_threshold_nonspam from -2 to -8.5 on my systems, but I'll have to lower it again soon because manually learning daily spam isn't enough... > Problem is - it *definately* affects accuracy. 2.5 has definately been > doing downhill since I installed it, with respect to how much spam gets > through. I was originally going to propose that the autolearn not > actually autolearn when nham >> nspam, but then it'd be difficult to > track changing trends in email. So I don't know what to do. I second that request : autolearning a ham should be active only when nham < 2*nspam || nham < 200 and respectively for a spam only when nspam < 2*nham || nspam < 200 For change tracking it should be the expiration's duty and in my case spam changes should be more often than ham changes. Denis Ducamp. -- http://www.groar.org/enough/bushit.jpg ------------------------------------------------------- This SF.net email is sponsored by: ValueWeb: Dedicated Hosting for just $79/mo with 500 GB of bandwidth! No other company gives more support or power for your dedicated server http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/ _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
