hi we used dspam_train for training. spam mails were collected in some random catchall account which had several thousands of emails.
we also did the same [EMAIL PROTECTED] ~]# /usr/local/dspam/bin/dspam_stats -H globaluser globaluser: TP True Positives: 13417 TN True Negatives: 5478 FP False Positives: 27 FN False Negatives: 571 SC Spam Corpusfed: 619 NC Nonspam Corpusfed: 0 TL Training Left: 0 SHR Spam Hit Rate 95.92% HSR Ham Strike Rate: 0.49% OCA Overall Accuracy: 96.93% we use classfication method ie globalgroup:classification:*globaluser where globaluser is trained with the spam and ham emails but there are some bugs related to this which we are already discussing in the forum with subject Re: [dspam-users] DSPAM - tagging question also please see this http://dspam.nuclearelephant.com/dspam-users/5413.html using merged groups -- well i have not used that yet. But please post your results online. we are planning to create a good readme file for dspam based on people's experiences, primarily so that hundreds of man-hours are not wasted in "re-inventing the wheel" in just trying to understand the current Readme.txt rather than understanding DSPAM. Your results will be useful for this also one other post mr. tonni is using shared groups successfully so all of this will add up to the common good rajesh mahadevan ---------- Original Message ---------------------------------- From: "Berger Stefan" <[EMAIL PROTECTED]> Date: Tue, 27 Mar 2007 16:27:18 +0200 >Hi all , > >I'm new to dspam and i'm a little confused about group settings . > >I'm running dspam ( 3.6.8 )in a clustered enviroment with a central >Mysql >Server ( write actions ) which is replicating to the local slaves . > >Dspam is running in daemon mode and is called via a shell script >from qmail-ldap . Messages were delivered to a central >NFS Server - with maildrop - and were stored in maildir format . ( I'm >not using the the >CGI ) > >Everything is working fine - Messages were delivered and dspam signature >is added to the message so customer can retrain by sending to their >alias adresses ( spam-user or nospam-user ). >Each User has is own dspam-user. > >Retraining is done with > >Sed '^X-DSPAM-/d' | dspamc --user user --class=innocent --source=error > >And > >Sed '^X-DSPAM-/d' | dspamc --user user --class=spam --source=error > > >Now i want to add a global user which should be trained by a Maildir >which >is feeded due a honeypot and good mails which we have collected in >another >Maildir . ( at time about 14000 Spam-mails and 4000 Ham-mails ) > >Can I feed the global user with these messages via dspam_train ? > >I have created a group file with following entry ( hoenymoon is my >global user ) > >honeymoon:merged:* > >In the logs i can see something like "user merged" but the results are >really bad . >Without the group file it's working better but the spam-hit rate is only >about 60 percent . > >Any Hints ? > >-Stefan >
