[Spambayes] Demand for support

Tony Meyer tameyer at ihug.co.nz
Tue May 11 04:05:56 EDT 2004


> Effectively you're right and French people often do the 
> mistake of using demand instead of request because it's so
> close to our French "demande"which has no impolite meaning.
> Sorry !

No need to apologise.  I figured it was something like that, but thought
maybe I should warn you since I can see that it would be easy for someone to
(mistakenly) take offense.

> I must apologize also about the fact that the reason of those 
> problems was completely on my side. The strange behaviour(s)
> I observed came from the size of my "pst" ( almost 2GB!) file
> and my trouble came from the fact that I installed SpamBayes
> 2 days before.
> 
> I'll try not to disturb you again with such problems.

No worries; glad that it's all going right now.

> 1-When SpamBayes automatically classifies mails in "Junk 
> E-mail" folder, may I consider it improved its knowledge of
> Spam with each of the mails classified there ? 

No, it does not do this, deliberately.  Since it already knows how to
classify those messages correctly, there isn't any need to train on those
messages.  (See also the comments below).

> 2-Same question with the mails claissified as "Junk suspects" 
> after I select them and click on "Delete as Spam" ?

Yes, when you click "Delete as Spam" SpamBayes updates its database
(knowledge) using that message.  (The same applies to the "Recover from
Spam" button).

> 3-SpamBayes Manager indicates to me : "You have much more 
> spam than ham. SpamBayes works best with approximately even
> numbers of Spam and ham". It's true that we mainly receive
> spam, and that's precisely why we're evaluating
> SpamBayes. Do you have any comments about it ?

You'll get better results by not simply training on everything that arrives.
Although it might seem that the more you feed SpamBayes the better things
will be, this is not actually the case - the 'hand waving' explanation is
that the real clues are drowned out by inconsequential ones.

If you only train on any mistakes that SpamBayes makes, and all mail that
ends up in "Junk Suspects", that should work well.  You'll probably still
end up training more spam than ham, but the ratio won't be as bad as if you
simply trained on everything.  Lowering the 'spam threshold' from 90% to 80%
would also help this.

This 'mistake based training' is really what the Outlook plug-in is designed
for, and you'll find it the most convenient, as well as (probably)
effective.

=Tony Meyer

---
Please always include the list (spambayes at python.org) in your replies
(reply-all), and please don't send me personal mail about SpamBayes. This
way, you get everyone's help, and avoid a lack of replies when I'm busy.




More information about the Spambayes mailing list