Author Topic: Problems with Bayes-it plugin (Read 32230 times)

mrakomor · « **on:** April 27, 2004, 09:19:35 AM »

Since today TheBat (version 2.10.01) Bayes it filter started to filter all incomming mails as spam. Could anyone give me a clue how to solve this problem? (I disabled it for the moment, however I would like to have the antispam plugin enabled). Thanks. :|

Anne · « **Reply #1 on:** April 29, 2004, 05:46:48 PM »

Seems to be a known problem with the Bayesit plugin. The Bayesit dev wrote this:

Quote

The only bug was located not in the filter, but in wrong copy of advanced.ini" file. This file is included into distribution of bayesit and it is placed in it's program folders. During first start it makes it's copy in the bayesit's working folder (so, every user can have his own settings). So, there are two copies of advanced.ini - one in the working folder (original, with comments), seconcd - in the current working folder (short copy, without comments). If the working folder and the program folder are the same places, it will be only one copy at all.

The problem is that due to some issues such copy created with wrong parameters inside. So, to resolve the problem it is necessary to exit The Bat! and then either manually copy the file "advanced.ini" from program folder to working folder, either (if these paths are same) open the distribution using winrar and owerwrite existing advanced.ini by one from the distribution.

The critical parameter in advanced.ini is "regarding threshold". If it is equal to zero, all mail will go to the "junk folder". The "normal" value is non-zero (usually 1.5..2.0).

One suggestion made to me recently was to try a different Bayes plugin by Achim Winkler which can be found here: http://www.lkcc.org/achim/download/bayesfilter.zip

davidc · « **Reply #2 on:** April 29, 2004, 07:25:34 PM »

here is the link for the latest bayesit
http://www.ritlabs.com/kb/idx/41/064/article/
download 0.53.rar
delete bayesit plugin
find your bayesit folder and overwrite with the contents of the .rar
(advanced.ini, bayesit.lng, bayesit.tbp, history.txt, kludges.txt
find tbplugins.ini (main folder of bat) and find section called plugins data.
find the bayesit line (number) and delete everything after the "=" sign.
add new bayesit plugin and I hope that the token option now defaults to a reasonble figure (mine defaulted to 15)
you will have to retrain but will have "white list"
it is easier than my words appear - sorry
:wink:

mrakomor · « **Reply #3 on:** April 30, 2004, 08:49:58 AM »

Thanks for all replies. I aplied suggested steps and plugin appears to be working fine.

However I have another question(s). I tried to explore the plugin settings and there is a thing I don't understand. I don't see the sense of field "Number of ranging tokens" in "Basic filter's options" (in the settings of plugin).

I was not able to find the user documentation for the plugin. If someone knows about it, I would be happy if he/she could give me the URL. It may be English, Czech or German.

davidc · « **Reply #4 on:** April 30, 2004, 04:14:53 PM »

I have not been able to find any readme or documentation, I have had to gather bits here and there and try out the options.

It was stated that due to the Bat installer (2.10) the value of the 'ranking tokens' was fixed at 0 and this caused all mail to be classified as junk. The value should be higher.
So I guess that the setting is an indicator of the strictness of the filter, and the users confidence in the training.
So I left the setting at 15 for training and then reduced it in steps, as bayesit compiled its base of spam and non-spam.
Seemed to work.

O hell, they have just released bayesit 0.54
:!:

davidc · « **Reply #5 on:** April 30, 2004, 04:19:44 PM »

And The Bat 2.10.03
I think I will take the advice given by Anne and wait for things to settle down.
:lol:

davidc · « **Reply #6 on:** May 01, 2004, 03:56:38 PM »

mrakomor
Pls ignore my comment about 'ranking token' value and leave at 15 (default)
Bad translation
I got confused with the advanced.ini file spelling of sighificant tokens. Note the 'h' instead of 'n'
This is in the main bayesit folder.
pls do not change
0.54 is mainly concerned with Secure Bat.

VenimK · « **Reply #7 on:** May 24, 2004, 05:08:46 PM »

Can this Bayes be tested one way or another ???

mrakomor · « **Reply #8 on:** May 25, 2004, 06:55:54 AM »

Sure, it can be tested. It requires however learning. You need a lot of spam for initial learning (you may ask someone around, he/she could send you backup of SPAM folders). Then you can try to send one of the messages to your address and watch if it is successfully moved to spam folder or remains at inbox.

steverio · « **Reply #9 on:** December 28, 2004, 06:12:28 AM »

This might be of use for someone...

From Ritlab's Forum:
Ritlabs

Quote

Marck Pearlstone
Administrator
Posted 12/21/2004 21:05:50

"Bayes" is a learning system.

Once you have marked a certain number of spams, it will start to get better at detecting them. I find that 1000 is a good number here.

You must also train it to recognise "ham" (non-spam) in a similar way (mark as non-spam).

I you find that it is not improving, close The Bat and use Windows explorer to delete the 'over-enthusiastic' dictionary. If too much is being passed as "non-spam", delete nspamdict.bye. If too much is classed as spam, delete spamdict.bye.

Having done that, start up TB and immediately retrain it with a good sample set - if it's spam, your junk folder should have a good sample set by now.

//.arck