Hi Jeff,
So if I wanted to merge any future training data for, say, foo and bar
into globaluser so that I can then keep globaluser up to date without
having to go in and manually feed it mail, and use it as a
classification source, I could do:
globaluser:merged:foo,bar
global:classification:*globaluser
And if I want to give globaluser the training data accumulated in the
past by foo and bar, I'd use dspam_merge to merge those dictionaries
together?
Regards,
Adrian
LedHed wrote:
Adrian Overbury wrote:
Hi,
I could be getting all this wrong, but I'm not sure if I'm
understanding how I'm supposed to set these groups up properly.
What I want to achieve in my environment is this:
We have a global dictionary that is a classification source of last
resort for anyone -- if they haven't got enough data or the filter
isn't too certain about the result it's coming up with. We'll call
this group global-class. If a new user comes along, and they don't
have enough tokens in their dictionary, dspam will consult global-
class for an answer as to the class of the message.
Three users in the system (call them foo, bar and quux) will be the
ones who'll feed global-class. They've already got data in their
dictionaries, so I'd be using dspam_merge to merge all this into
global-class. For every message that any of these three accounts
recieve, the tokens extracted from it will go into global-class'
dictionary, and the signature will belong to global-class (so that
if it's misclassified, and they forward it to [EMAIL PROTECTED] or
[EMAIL PROTECTED] to retrain it, it relearns it in global-class'
dictionary).
Now, to achieve all of this, I think that what I would need to have
in $DSPAM_HOME/group would be:
global-class:merged:foo,bar,quux
global-class:classification:*
And the merge command line would be:
% dspam_merge foo bar quux -o global-class
And the stuff with the signatures and relearning should Just Work
(assuming I already have these working in my system, which I do),
or am I missing something here? I'm not sure.
Regards,
Adrian
Adrian,
I don't think its necessary to call dspam merge. In a merged group
untrained or inadequately trained users will use the tokens from
foo, bar, & quux until their own personal training has matured.
To achieve this you would add this to your group file.
trainers:shared:foo,bar,quux # Merged Group Trainers
trainers:merged:* # Merged Group
It may also be possible to just comma separate the trainers and omit
the shared group "trainers"
foo,bar,quux:merged:* # Merged Group w/ foo, bar,
quux as base users (trainers)
Personally I've never tried using a group of users as the base user
for a merged group, I've only ever used a single user.
In theory this should work.
Hope this helps.
-Jeff Harris
