[spambayes-bugs] [ spambayes-Feature Requests-1030515 ] dealing with *much* more spam than ham

[SourceForge.net]

Feature Requests item #1030515, was opened at 2004-09-19 12:33
Message generated for change (Comment added) made by anadelonbrin
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=498106&aid=1030515&group_id=61702

Category: Interface Improvements (example)
Group: None
>Status: Closed
Priority: 5
Submitted By: ajki (ajki)
Assigned to: Nobody/Anonymous (nobody)
Summary: dealing with *much* more spam than ham

Initial Comment:
"Warning: you have much more spam than ham - SpamBayes
works best with approximately even numbers of ham and
spam."

That's a warning msg from Spambayes (and I'm getting
poorer results in the last weeks).

My Spam/Ham-ration is quite normal, I guess: about
75:25 for the last two years. I think Spambayes should
cope with reality and do not assume a false 50:50
percent ratio.

Best,

Andreas


----------------------------------------------------------------------

>Comment By: Tony Meyer (anadelonbrin)
Date: 2004-09-20 13:34

Message:
Logged In: YES 
user_id=552329

1.  If you are getting satisfactory results, then you
obviously don't need to take notice of the warning.

2.  There's a large difference between the ratio of
ham::spam received and the ratio trained.  Different
training techniques (see the wiki for details) result in
quite different ratios.

3.  It's unlikely there is a "normal" ratio.  Some people
get much more ham than spam (those who receive several high
volume mailing lists, for example, or those who receive very
little mail at all, and whose address is hardly used). 
Others, like you, get much more spam than ham.  Some,
actually do get about 50:50.  SpamBayes needs to be able to
cope with all of these.

4.  If you can suggest improvements to the math that will
help this, then please do!  This is open-source, and that's
what it's all about.

I'm closing this, as this is a recognised issue to be worked
on for 1.1 in several places, and we don't really need
another open tracker about it.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=498106&aid=1030515&group_id=61702