spamassassin

[Maxime Delorme]

I've got problem,

spamassassin doen't work properly. It is trained but I think it doesn't 
use it bayesian teaching. Because a lot of same spam learned are not 
identify as spam when they come back.

On kolab server I've got :

/root/.spamassassin/bayes_toks
/root/.spamassassin/bayes_seen

updated with spamanclam.sh

But also
/usr/local/kolab/.spamassassin/bayes_toks
/usr/local/kolab/.spamassassin/bayes_seen
/usr/local/kolab/var/amavisd/.spamassassin/bayes_toks
/usr/local/kolab/var/amavisd/.spamassassin/bayes_seen

So I'm thinking, that spamd  is not using the good one bayes_toks, how 
could I know

Any help would be great

Thank everybody