On Wed, Dec 10, 2003 at 08:55:16PM -0600, Benjamin Story wrote: > Hi all, > > I've been trying to add the Bayesian part of SA to my company's mail > gateway by having users put spam and ham into an exchange public > folder from which I have a perl script downloading the messages into a > directory for sa-learn. My problem is that everytime I run sa-learn > it processes the messages and says 0 learned. When I do a sa-learn > --dump to see the bayesian table nothing is output. Any pointers > would be greatly appreciated.
For my setup, I've just used on each file: sa-learn --spam --single < file or sa-learn --ham --single < file I'm using 2.55, and it doesn't have a --dump flag, so I can't compare. If you have autolearned files, you would get the behavior above since the system already knows them. If you have successfully learned them before, you would also get it. So if you are running the same directory over and over and wondering at it, they might just be learned already. You might try sa-learn --forget --dir *directory* and then try to relearn it. If that doesn't unstick something, you might try -D for debugging messages. Maybe you don't have write permission to the database you are trying to create? I had to make sure my database files were owned by the "mail" user in my setup (and were in that home directory). I guess I'd also recommend looking through the script. It's about 48 lines (plus the perldoc documentation). It probably passes on to utility scripts, but you should be successful in sorting things out. Of course you'll understand the software a good bit better when you are done too. The refuge of the desparate is strace, but I've often seen that be successful in teasing things out. You'll be able to see what files are being opened and such, which can often give you a clue. Sometimes its easier than looking at the source and sometimes not. -- Don Bindner <[EMAIL PROTECTED]> ----------------------------------------------------------------- To get off this list, send email to [EMAIL PROTECTED] with Subject: unsubscribe -----------------------------------------------------------------