Michael Moncur wrote: MM> > Ok, I just checked in a "fixed" version of Daniel's suggested MM> > change. It's not MM> > doing the adjustment-to-threshold-based-on-message length SO BE MM> > WARY IF YOU'RE MM> > IN THE HABIT OF USING CVS -- THIS CURRENT CODE MIGHT YIELD A MM> > BUNCH OF FALSE MM> > POSITIVES. It's somewhat unlikely, I think it's still a pretty MM> > tight rule, but MM> > be careful for a day or two. MM> MM> Actually it seems harmless - unlike the old spam phrases stuff, there's MM> still only one rule and PORN_3 has a score of 0.6, so it's not going to push MM> too many things over the threshold.
Yes; I wanted to make sure to let people know that we've done something that is likely to increase the rate of false positives, however insubstantial that rise will be. Generally, I think all agree that we want to stay very focussed on keeping the false positive rate as low as technically possible. We can apply further technique to the current system to reduce the false positive rate, so this is an ephemeral condition. MM> Perhaps after testing it might be good to have a separate LOTS_OF_PORN_3 MM> rule that checks for a higher number... Yes, actually I think maybe the way to do it is almost exactly like the spam phrases, where we can pass a parameter to the eval function, and have not just two, but maybe 3-4 levels of porniness. C _______________________________________________________________ Have big pipes? SourceForge.net is looking for download mirrors. We supply the hardware. You get the recognition. Email Us: [EMAIL PROTECTED] _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
