On 25.3.2010 10:14, Giampaolo Tomassoni wrote:
> Having this in the bayes db:
> 
>       335424 nspam
>       144915 nham
>       129892 ntokens
> 
> and a fairly good hit rate by Bayes in detecting both spam and ham, how
> would you improve Bayes scores?
> 
> In example, would you increase every bayes scores by a fixed percentage, or
> instead would you focus on score borders (definitely ham or spam, ie:
> BAYES_00, BAYES_05 and BAYES_95, BAYES_99)?
> 

I have increased BAYES_00 and BAYES_99. It seems that those are pretty
good and cause no FP's, but BAYES_05 may sometimes be spam. I have
BAYES_99 as a killer, it has 5 points, sending the mail to a 'probable
spam' alone.

-- 
http://www.iki.fi/jarif/

You will never know hunger.

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to