[dspam-users] dspam_train method / shared or merged ?

Hi,

im training my dspam using dspam_train ,

now I collected a bunch of mails using postfix always_bcc options to a local
account on my dspam box
and used procmail and spamassassin to filter spam from ham into different
folders..

now I collect about 9000 ham messages and 9000 spam messages..

I feed this into this using the following command dspam_train galileogroup
/tmp/dspam/spam /tmp/dspam/ham

Now once it has finished trained, the spam statistics goes up 9000 for good
then spam messages but then
the hit rate on spam messages the next hour is terrable.. like 17 per hour?

what could be the problem?

Iv got 9000 users on my system and im using a shared group, what would be
the best group for this solution
as I dont want users to train the dictionary ..

I notice when I use a shared group, TP True Positives: TN True
Negatives: automaticlly increment with out trainig,
I dont understat whats going on there? could someone explain?

So should I be using a merged or shareg group?
Received on Mon Oct 23 02:45:01 2006