bogotune and "exhaustion"

[David Relson]

On Mon, 29 Mar 2004 07:02:20 -0500
Tom Allison wrote:

> Ran into a cute catch-22.
> 
> bogotune wants a sample size that includes some high scoring ham and
> low scoring spam (maybe) to get a good calculation of what to set the 
> parameters at.
> 
> run corrections to exhaustion tends to remove that high scoring ham, 
> giving you a big fat remark and a shortened bogotune output.
> 
> So, it seems that I can do one or the other but not both on my
> archives. Or I'm doing something wrong.

Hi Tom,

What you say sounds reasonable.  Bogotune needs the variation in ham
scores so that it can adjust the number of false positives.  I can
believe that train-to-exhaustion conflicts with this.  It's also
possible that bogotune could be modified to not need the high-scoring
ham.  This possibility will need some thought since a different way of
selecting spam_cutoff will be needed.  I'll think about it, but can't
guarantee a solution.

David