Hello Dspam list, I'm currently testing a setup of dspam 3.9.1 and, after a few weeks of running it, it seems that learning doesn't happen. I have the same X-DSPAM-Confidence: 0.6589 and X-DSPAM-Probability: 0.3411 over and over again...
I first assumed that, considering my low ratio of spam (around 1.5%, greylisting is working well already), it was normal for Dspam to take some time to populate the dictionary. But now, after testing the same email several time and not seeing any change in the probability, I tend to doubt my configuration. Here are my statistics: ---- # dspam_stats jul...@linuxwall.info jul...@linuxwall.info TP: 0 TN: 4175 FP: 0 FN: 47 SC: 0 NC: 0 # dspam_dump jul...@linuxwall.info |wc -l 2226565 # dspam_dump jul...@linuxwall.info |head 16562406131549251947 S: 00000 I: 00001 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 6865589383638280368 S: 00000 I: 00001 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 9170080104461015804 S: 00000 I: 00003 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 5926792591516926582 S: 00000 I: 00003 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 14071509996661893476 S: 00000 I: 00001 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 12628252971455951009 S: 00000 I: 00022 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 8466811669071571900 S: 00000 I: 00003 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 5005764671929148649 S: 00000 I: 00001 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 15031205318376739124 S: 00000 I: 00001 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 7946915292529287299 S: 00000 I: 00001 P: 0.5000 LH: Fri Sep 24 12:00:31 2010 terminated. ---- If I send myself a spam email contain the token 'viagra+spam', I receive the email properly with confidence 0.6589 and probability 0.3411. The token exist and is marked as innocent: ---- # dspam_dump jul...@linuxwall.info viagra+spam 3709864854363566654 S: 00000 I: 00001 P: 0.5000 ---- Now, I relearn this email as spam using the web interface. The token is now marked as spam ---- # dspam_dump jul...@linuxwall.info viagra+spam 3709864854363566654 S: 00001 I: 00000 P: 0.5000 ---- Why hasn't the probability changed ? I resend the exact same email one more time. The token is updated, but marking it as spam one more time doesn't change anything... ---- # dspam_dump jul...@linuxwall.info viagra+spam 3709864854363566654 S: 00001 I: 00001 P: 0.5000 === mark as spam === # dspam_dump jul...@linuxwall.info viagra+spam 3709864854363566654 S: 00002 I: 00000 P: 0.5000 ---- The configuration is here: http://jve.linuxwall.info/dump/dspam.conf.html (or without the .html for the text file). Did I do something wrong ? Is it due to using 'markov' for PValue ? Should I switch back to bcr ? Thanks, Julien ------------------------------------------------------------------------------ Nokia and AT&T present the 2010 Calling All Innovators-North America contest Create new apps & games for the Nokia N8 for consumers in U.S. and Canada $10 million total in prizes - $4M cash, 500 devices, nearly $6M in marketing Develop with Nokia Qt SDK, Web Runtime, or Java and Publish to Ovi Store http://p.sf.net/sfu/nokia-dev2dev _______________________________________________ Dspam-user mailing list Dspam-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dspam-user
