Hello Thorsten, i sorted out to get dspam working, but it tags too much mails as spam, even mials that are not spam,
now thats the poinzt where dovecot-antispam comes in, now my idea is, that mails that get "spam" tag by dspam, to retrain by "dovecot-antispam". but i dont get thru it. maybe this is now the pont where you can help me ?
marko
Thorsten Vollmer schrieb:
On Fri, 2008-11-07 at 13:31 +0100, Marko Weber wrote:
Thorsten Vollmer schrieb:
On Thu, 2008-11-06 at 17:03 +0100, Marko Weber wrote:
Ok i summarized some things, maybe any1 can give a comment to it =
when i restart dovecot i get this =
/etc/init.d/dovecot restart Warning: Growing pool 'settings' with: 8192
- Stopping dovecot ...
[ ok ] Warning: Growing pool 'settings' with: 8192- Starting dovecot ... Warning: Growing pool 'settings' with: 8192 ILoading modules from directory: /usr/lib64/dovecot/imap WGrowing pool 'module loader' with: 2048 IModule loaded: /usr/lib64/dovecot/imap/lib90_antispam_plugin.so IEffective uid=65534, gid=65534, home= WGrowing pool 'antispam-pool' with: 2048
[ ok ]email-headers of SPAM tagged mails (spamassassin + pyzor + razor detect them as spam cause they are spam) dspam tags the mails like this =
X-DSPAM RESULT Innocent X-DSPAM Confidence 0.8505 X-DSPAM Probability 0.0000
so, Spamassassin detect them as spam and dspam not ? (to an old mail in list question , yes i trained ca 10.000 spam mails against some not spam).
because this mail would grown in endless long text i pasted the configs on web =
POSTFIX main.cf = http://rafb.net/p/PQy5FN10.html POSTFIX master.cf = http://rafb.net/p/2l02A087.html DOVECOT dovecot.conf = http://rafb.net/p/PjvV1h44.html DSPAM dspam.conf = http://rafb.net/p/7O8lEf46.html
I looked only briefly at the files: You need to set "signatureLocation=headers", but this is unrelated to your current problem.
I use Postfix with "virtual users & domains"
dspam.debug shows this entries when i send a mail to the server = http://rafb.net/p/sbEqK070.html
Sending an actual spam message would have been more useful.
i forwarded an spam mail to the server with dspam.
Do the results improve if you disable the "naive" statistical algorithm?
You may also try Markovian weighting, but you need to disable ProcessorBias in that case.
Not all combinations of Tokenizer, PValue, Algorithm and ProcessorBias result in working setups. Try other ones. But remember that you need to rebuild the corpus if you change the tokenizer.
sorry, guys. but how do i get dspam working 100% , and how do i get this retraining working ? where i can move mails out spam into incoming and other way spam from incoming to spamfolder to retrain dspam.
yes, maybe alot stuff, but any feel free to help or to explain ?
thanks ahead from hamburg
marko