[Spambayes] Promoting Spambayes (was Re: FYI: Java implementation)

[Phillip J. Eby]

At 10:28 AM 1/21/03 +0000, Michael Hudson wrote:
>Tim Peters <tim.one at comcast.net> writes:
> > Anthony Baxter has some plots of score distributions for Graham-combining,
> > Gary-combining and chi-combining here:
> >
> >     http://spambayes.sourceforge.net/background.html
>
>I meant to say it when I first looked at that page, but seeing those
>plots nearly made my eyeballs fall out.  Why does anyone still use
>Graham-combining?

Because nobody's seen the plots, obviously.  :)

I think what would be needed to change that would be:

1. A Spambayes release

2. A "spam shootout" wherein half a dozen Bayesian mail filters (e.g. 
Popfile, Mozilla, other...?)  are tested against the same corpus, using the 
cross-validation testing mechanism.

3a. Spambayes comes out on top, with a fraction of the error rate of 
others: publish the results, get a Slashdot story, and slashdot the project 
site.  :)

3b. Spambayes doesn't come out on top: find out why, fix the problem, go to 
step 3a.  :)