Re: Too many false positives...goodbye, razor

Nice, those numbers are quite informative Theo. Makes me think that I was 
right to set my min_cf to 11, effectively eliminating the noise present in 
range_01_10, which seems to have a pretty bad S/O.

It also makes me think that upping my min_cf past 11 is also well worth it.

Thanks for posting the data Theo.



At 06:38 PM 2/11/2003 -0500, Theo Van Dinter wrote:
> On Tue, Feb 11, 2003 at 06:30:46PM -0500, Matt Kettler wrote:
> > In general it seems to be rare that it FPs when the cf is very high, but
> > there's a LOT of flaky matches in the lower end.
>
> Well, based on ~90k messages (wow, that's not many...) that are from a
> recent spamassassin mass-check run:
>
> OVERALL%   SPAM%     HAM%     S/O    RANK   SCORE  NAME
>  16.286  42.8416   0.1639    0.996   0.98    0.01  RAZOR2_CF_RANGE_91_100
>   1.095   2.7929   0.0641    0.978   0.87    0.01  RAZOR2_CF_RANGE_21_30
>   0.205   0.5222   0.0125    0.977   0.87    0.01  RAZOR2_CF_RANGE_81_90
>   0.268   0.6777   0.0196    0.972   0.86    0.01  RAZOR2_CF_RANGE_41_50
>   0.145   0.3667   0.0107    0.972   0.85    0.01  RAZOR2_CF_RANGE_51_60
>   0.806   1.9686   0.0997    0.952   0.81    0.01  RAZOR2_CF_RANGE_11_20
>   0.165   0.3990   0.0232    0.945   0.79    0.01  RAZOR2_CF_RANGE_61_70
>   0.173   0.4137   0.0267    0.939   0.77    0.01  RAZOR2_CF_RANGE_31_40
>   0.553   1.2175   0.1496    0.891   0.66    0.01  RAZOR2_CF_RANGE_01_10
>   0.147   0.3227   0.0410    0.887   0.65    0.01  RAZOR2_CF_RANGE_71_80
>
> So basically what this means is that all CF values have false positives.
> 91-100 is the best matching the most messages while not FPing too much.
> Everything else seems to be more questionable due to low hit rates.



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf