I think a lot of work has been done to reduce false positives, and often
this may have the effect of increasing false-negatives (eg. badly
implemented AWL).  I agree that it might have gone a bit far and that more
spam does seem to be slipping through.  Luckily, it's fairly easy to rectify
this by changing a parameter in the GA score evolver, which currently treats
false-positives as "20 times worse" in some sense than false-negatives when
it's doing its optimization.  I'm currently working on getting a non-spam
corpus together so I can run the GA in justin's absence -- hopefully that'll
get up and ready within the next week or so.

Having said that, I think apart from the issues with AWL, it's not *too*
bad.

C

on 2/1/02 6:16 AM, CertaintyTech - Ed Henderson at
[EMAIL PROTECTED] wrote:

> So far I have seen the following with v2.01 as compared to v1.5:
> 1.  2.01 appears to be better at reducing false positives that v1.5 - this
> is good!
> 2.  2.01 appears to be worse with false negatives.  There is alot more Spam
> getting thru.  I don't have the ability to run the false negatives thru 1.5
> but I'm betting it would have caught some of them.
> 
> I realize that 2.01 was a significant change from 1.5.  Was there a greater
> emphasis on reducing false positives?  Is what I'm seeing typical for other
> SA users out there?
> 
> Thanks,
> Ed.


_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk

Reply via email to