On Wednesday 30 August 2006 00:17, Raúl Sánchez Siles wrote: > Hello: > > <intro> > After some thiking I finally decided to move from Thunderbird to Kmail. > This was some months ago. After some problems with the mail solved by > kmailcvt and others[1] I made kmail work with no problems. > > Then I tried to configure spamassasin so that kmail could manage spam. > After testing some time, I noticed that I was quite happy with the TB > simple bayes spam filter. I know that there are also some simple bayes > filter that work with kmail (e.g.: bogofilter) but I thought that since SA > is well known to be the best and the most used spam filter I decided for > it. > </intro> > > I have it configured and runs quite well, but it seems for me that the > bayes training (sa-learn) is not working properly. As the wizard defines I > have a rule for spam and another for ham. both at the end of the filter > pipeline so they are never run automatically, instead of that I have some > icons on the main bar for that. > > The spam rule action is this: sa-learn -L --spam --no-sync > The ham rule action is this: sa-learn -L --ham --no-sync > > After some training I have the impression that this is not working > properly and my suspicion is that kmail is not passing the whole e-mail to > sa-learn so that it could learn the tokens.
The reason is the --no-sync option which stores what is learned outside the main spamassassin database. You need to manually run 'sa-learn --sync' from time to time, to incorporate the learned results into the spamassassin database. You could instead remove the --no-sync options, but that would slow down the spam filtering operations so is not a good idea if you receive much mail. -- David Jarvie.
