On Feb 27, 2013, at 11:17 AM, Stevan Bajić wrote: > Am 2013-02-27 17:30, schrieb Chad M Stewart: >> On Feb 27, 2013, at 7:38 AM, Stevan Bajić wrote: >> >>> It does not look like you have made anything incorrect. 91.5% >>> accuracy >>> slightly after the training phase is not bad. What tokenizer are you >>> using? >> >> Tokenizer osb >> >> > You just got recently out of training phase and 91% is not bad for > that. The number will get soon higher. The reason for it to be around > 90% slightly after the training period is that DSPAM was in a learning > phase and it is normal to make errors in that phase. Later on this > number will get higher and higher because DSPAM will (usually) make less > errors and whitelisting will help to keep the FP rate lower. I would not > break my head because of the 91%. Let DSPAM run for a bunch of > days/weeks and if you then still have the feeling that it is not > accurate enough then it's time to act. But not now.
Cool. I can do that. With the number of mailing lists I'm on, hitting the 2500 mark for training does not take long. I don't mind messing with my account if the end result is a better overall experience for all of my users. Thankfully the worst list for spam (misc@openbsd) has gone to a member only model, so no more spam from there. On the flip side when I was doing business with equifax I gave them a unique-to-them-only address, the amount of spam to the address is ridiculous. Of course that's after my pre-dspam filters. -Chad ------------------------------------------------------------------------------ Everyone hates slow websites. So do we. Make your web apps faster with AppDynamics Download AppDynamics Lite for free today: http://p.sf.net/sfu/appdyn_d2d_feb _______________________________________________ Dspam-user mailing list Dspam-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dspam-user
