On the other hand, if I auto-learn Bayes, then the Bayes database gets very
big, and any mistakes that SpamAssassin does, and that I manually correct by
sa-learning them to Bayes, don't get much weight. Correct? So fine-tuning
Bayes gets harder if I use auto-learn?

Also if I auto-learn this mailing list (because everything goes through SA)
then Bayes may get confused about these "penis enlargement" messages. OK,
only if score is below -2, or if score is higher than 15, if I use the
default settings. That must be the reason why these settings are so high?

----- Original Message ----- 
From: "Matt Kettler" <[EMAIL PROTECTED]>
To: "Harri Pesonen" <[EMAIL PROTECTED]>;
<[EMAIL PROTECTED]>
Sent: Sunday, August 03, 2003 7:16 PM
Subject: Re: [SAtalk] The correct use of Bayes


> At 05:02 PM 8/3/03 +0300, Harri Pesonen wrote:
> >I think that the autolearn is bad, because it only learns from those spam
> >messages that it knows to be spam for certain in any case, and the same
> >for ham as well. OK, it does not necessarily mean that Bayes agrees what
> >SpamAssassin thinks. Anyway, currently I have autolearn disabled and only
> >feed Bayes those messages that SpamAssassin misses, spam or ham. Is this
a
> >correct way of thinking?
>
> To conclude that autolearn is bad is a considerable misunderstanding of
> bayes. Bayes does NOT learn emails. It learns tokens. What bayes learns
> from one spam is applicable to other spams. Bayes can apply these lessons
> to find spams that would otherwise miss the ruleset, even if it's only fed
> 'high scoring spam".
>
> In theory you should feed your bayes engine a fairly balanced diet of spam
> and nonspam, without consideration of wether or not SA caught it.
>
> >
> >Of course I had to sa-learn a minimum of 200 messages before Bayes
started
> >working. Could it be that the Bayes stops working correctly if autolearn
> >is not enabled?
>
> It should continue to work even with autolearn off, but this does increase
> the need for you to keep feeding it messages manually and to feed it a
> balanced diet.
>
>
>



-------------------------------------------------------
This SF.Net email sponsored by: Free pre-built ASP.NET sites including
Data Reports, E-commerce, Portals, and Forums are available now.
Download today and enter to win an XBOX or Visual Studio .NET.
http://aspnet.click-url.com/go/psa00100003ave/direct;at.aspnet_072303_01/01
_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk

Reply via email to