On Thu Jul 03 13:39 +0200, Giuseppe Ghibò wrote:
- let spam pass to users (but with X-Spam-Status) and at the same time collect
all the recognized spam to a repository for further bayes learning.
Collecting SA mails to a bayesian system is beyond stupid, as far as I can tell. You go through all the computational expense of bayesian to essentially duplicate the SpamAssassin rules..
AFAIK the sa-learn is only performed once, that it can be useful for training a 2nd server, maybe testing different thresholds, IMHO.
BTW, has anyone tried the crm114 filter?
Bye. Giuseppe.