Training from scratch

[David Relson]

On Mon, 07 Mar 2005 20:14:40 +1000
Mark Constable wrote:

> Debian bogofilter v 0.93.5-1 using courier/maildroprc with
> a default /etc/bogofilter.cf except for "spam_cutoff= 0.80".
> 
> I thought I'd try training from empty to see how that goes
> and I'd like to check that I have this worked out right. All
> messages coming in go thru...
> 
> . xfilter "/usr/bin/bogofilter -u -e -p -d $HOME/.bogofilter"
> 
> then any spams in an IMAP "retraining" folder go thru...
> 
> . bogofilter -d${MPATH}.bogofilter -Ns < "$i"
> 
> and any good messages in an alternate retraining folder go thru...
> 
> . bogofilter -d${MPATH}.bogofilter -Sn < "$i"
> 
> Just now I tried 10 emails in a row from myself and they all
> ended up as Unsure with a weight of 0.52. I manually fed them
> thru the above rule and yet they continued to have a weight
> of 0.52. I would have assumed that after training about 200
> words 10 times via "-Sn" that bogofilter would start to register
> these tokens as "good" and the 0.52 weight would decrease ?
> 
> ($MPATH is each users homedir for mail, is is correct)
> 
> --markc

Hi Mark,

Sounds like something's broken.  There are a couple of things to do:

*** Run "bogoutil -d ${MPATH}.bogofilter/wordlist.db" to see that
training is actually happening.

"bogoutil -w ${MPATH}.bogofilter/wordlist.db .MSG_COUNT" will show the
message counts in your wordlist.  They should be changing over time.

*** Add logging "-l" to your bogofilter commands.  Use "-l" with
your first command and "-v -l" with the others.  This will add lines
like the following to syslog:

Mar  7 00:01:02 nic bogofilter[32267]: register-Ns, 161 words, 1 messages
Mar  6 20:07:57 nic bogofilter[27952]: X-Bogosity: Spam, spamicity=0.949556, version=0.94.0, register-s, 97 words, 1 messages 

*** Also you can use "bogofilter -vvv <msg" to see the spam and ham
counts and the spamicity scores for each token in the message.  The "-
vvv" output is described in the FAQ.

HTH,

David

_______________________________________________
Bogofilter mailing list
Bogofilter at bogofilter.org
http://www.bogofilter.org/mailman/listinfo/bogofilter