Sound advice. Thanks for the reply and input - both are greatly appreciated!
And thanks again to Ryan More and Matt Kettler. --Larry > -----Original Message----- > From: Giles Coochey > Sent: Thursday, October 30, 2003 6:55 AM > To: Larry Gilson; [EMAIL PROTECTED] > Subject: RE: [SAtalk] Bayes spam-ham or ham-spam ratios > > > Your bayes_auto_learn_threshold_nonspam should be a LOT lower > than 6. > > If you've actually been learning emails which score 5.9 as > non-spam then you've most likely poisoned your bayes > database, especially with 2.55 as many spams are now being > proliferated to deliberately score less than 5 with SA2.55 - > you should probably clear out your current db and re-run > sa-learn on your spam/ham corpus. > > Additionally bayes_auto_learn_threshold_spam is set rather > low (at 8) - you might learn non-spam as ham, giving you > further glitches. > > The defaults (0.01 for nonspam and 12 for spam) are not > arbitrary values, but values that have been found to > generally work well in many different areas, I wouldn't > bother trying to tweak them unless you find you have a problem. > > Remember: If it's not broken, don't fix it. > > -----Original Message----- > From: Larry Gilson > Sent: 28 October 2003 20:20 > To: [EMAIL PROTECTED] > Subject: [SAtalk] Bayes spam-ham or ham-spam ratios > > > I am using 2.55. If the answer is different for 2.60, please > answer with respect to 2.60 as I will be upgrading *very* soon. > > I am experimenting with Bayes in a site-wide/gateway > configuration. One thing that I believe is affecting my > tests is the external Procmail whitelist. I find the number > of spams learned to the number of hams learned extremely large. > > What is the ratio of ham:spam or spam:ham (2:1, 3:1) that > Bayes becomes inneffective? Does either ratio make a > difference? I am thinking in terms of numbers of messages > learned. Should I be thinking differently? > > Also, my local.cf has the following > > bayes_auto_learn_threshold_nonspam 6 > bayes_auto_learn_threshold_spam 8 > > Is this appropriate or did I misinterpret the > Mail::SpamAssassin::Conf doc incorrectly. > > Thanks, > Larry ------------------------------------------------------- This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/ _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
