[TriLUG] Re: more spamassassin, procmail, damaged 'from' line and beyond

Mark Shuford davemarcus at pobox.com
Tue Apr 15 22:23:34 EDT 2003


On Tue, 15 Apr 2003 17:02:20 -0400
James Manning <jmm at sublogic.com> wrote:

> 
> Also, you wanna keep around spam (and ham) for bayes learning.  You
> can set some auto-learn settings (or keep the defaults), but training
> uncaught spam is critical to SA getting better over time.
> 


Well, I don't know about SpamAssassin; I'm using SpamBayes. I've had that going since mid February and have not gotten a single false positive yet. I get quite a few false negatives but that's fine. SpamBayes appears to be rather conservative so it doesn't throw out good stuff -- or maybe it's that I trained it on about a year's worth of HAM with only about two weeks of SPAM.

I just set it (last night) to put SPAM directly in my xSPAM mailbox. This gets trained on every night (or early mourning (sic)) so anything that was false negative I droop in there. Then the xSPAM box gets emptied. Bye bye.

I look in xSPAM a couple times a day for false positives. These go into a folder called either xHAM (if it's to be filed for later) or xHAM-trash if it just needs to be re-trained as HAM after I've read it. At training time the xHAM boxes retrain SpamBayes on this HAM. Then xHAM-trash is emptied and xHAM messages are moved to xHAM-file for me to file away. (I guess I could set up some more automated stuff to file this stuff automaticly...)

(The 'x' exists to make all this file at the bottom of my folder lists... I guess I could use 'z'. I used to try to get stuff I wanted to file at the top there by preceding with '_' but this doesn't seem to work with all mailers. A few seem to ignore leading non-alpha characters.)


As I said: I don't know SpamAssassin. Is it Bayes Filtering based also?





-- 
Mark Shuford

Now I know what a statesman is; he's a dead politician. We need more statesmen.
					--Bob Edwards

Visit me at: http://www.krazeemark.com  (Got the blog up...)



More information about the TriLUG mailing list