https://issues.apache.org/SpamAssassin/show_bug.cgi?id=6247

--- Comment #55 from Justin Mason <[email protected]> 2009-12-21 05:19:51 UTC ---
(In reply to comment #54)
> > rescore masscheck logs with mcsnapshot + today's trunk Scores
> > HABEAS, BSP and SSC and DNSWL Disabled
> > ==========================
> > # SUMMARY for threshold 5.0:
> > # Correctly non-spam: 703899  99.93%
> > # Correctly spam:     2565063  98.49%
> > # False positives:       467  0.07%
> > # False negatives:     39257  1.51%
> > # TCR(l=50): 41.597904  SpamRecall: 98.493%  SpamPrec: 99.982%
> 
> So that means that 
> 
> 467-387
> --> 80    
> 80/(703899+467)
> --> 0.000113577316338381  
> 
>     => 0.01135% of hams were rescued from being FPs by the DNSWL rules.
> 
> 41888-39257
> --> 2631                
> 2631/(2565063+39257)
> --> 0.00101024451680285  
> 
>     => 0.101% of spams were, conversely, allowed through by them.

actually, I've realised this may be measuring the wrong thing -- since FP% is
much lower (by design) than FN%, it's quite likely that we'll see a
comparatively high rate of hits against spam, even if the rules are useful.  A
better way to think about it is in terms of those values compared against the
FP%/FN% rates.

  => 80 hams rescued from the 467 FPs = 17.13% of FPs rescued

  => 2631 spams added to the 39257 FNs = 6.70% additional FNs

when looked at this way, they look more practical.  a reliable way to cut the
FP rate by 17% is definitely useful, although obviously it'd be nice if the
6.7% additional FNs was reduced.

-- 
Configure bugmail: 
https://issues.apache.org/SpamAssassin/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

Reply via email to