I thought I'd mention that since I wiped and rebuilt my Bayes files my
success rate has jumped from 93% to 99+%. Considering the amount of spam
pouring into here, that's not bad at all.
----- Original Message -----
From: "Anthony Peacock" <[EMAIL PROTECTED]>
Cc: <users@spamassassin.apache.org>
Sent: Friday, January 26, 2007 9:49 AM
Subject: Re: Bayes
Hi Dan,
Dan Barker wrote:
-----Original Message-----
From: Anthony Peacock [mailto:[EMAIL PROTECTED]
Sent: Friday, January 26, 2007 7:22 AM
To: users@spamassassin.apache.org
Subject: Re: Bayes
<snip>
I also manually learn by mistake. All FNs & FPs are fed back to the
system. And I occasionally feed some recent ham as ham.
This has worked really well for me over a number of years.
</snip>
Anthony, I only train the False Negatives that do not already have the
BAYES_99 hit. Should I be training all FNs? It seems to me that there
could
not be any improvement by training these messages.
Please explain why this is wrong (if it is).
Others that know Bayes better may be able to give you the definitive
answer about whether or not it makes any difference.
Personally, I don't think it can do any harm, and it may also reinforce
the spammyness of the tokens.
It is also possible that the BAYES_99 hit came of some tokens in the
email, but there may be others that are not yet in the database, or not as
significant for spam. And feeding the email into the learning system will
increase these tokens spammyness. But this is pure speculation on my
part.
The bottom line for me is that it can't do any harm, it may do some good,
so I don't expend the effort to manually distinguish between those types.
--
Anthony Peacock
CHIME, Royal Free & University College Medical School
WWW: http://www.chime.ucl.ac.uk/~rmhiajp/
"If you have an apple and I have an apple and we exchange apples
then you and I will still each have one apple. But if you have an
idea and I have an idea and we exchange these ideas, then each of us
will have two ideas." -- George Bernard Shaw
