Understanding tuning results

[Greg Louis]

On 20030611 (Wed) at 1450:20 +0100, Peter Bishop wrote:
> On 11 Jun 2003 at 12:53, Boris 'pi' Piwinger wrote:
> 
> > > robs  min_dev spam_cutoff run0 run1 run2 total
> > > 1.000 0.475   0.503000     62   66   78  206
> > > 0.320 0.475   0.503000     62   66   78  206
> > > 0.100 0.475   0.503000     62   66   78  206
> > > 0.032 0.475   0.503000     62   66   78  206
> > > 0.010 0.475   0.503000     62   66   78  206
> 
> As a potential user of tuning.sh, the new format
> looks much better than the current one

There's no harm in changing tuning.sh's reporting format, but what gets
written into $PARM_TBL should not be altered unless you have no need
for data reduction.

As a heads-up: I'm developing a script called bogotune that will take
as input the names of some spam and some nonspam mbox or message-count
files, and will use them plus the training database to derive
suggestions for cache size, Robinson's x, Robinson's s, the minimum
deviation, the spam cutoff and the nonspam cutoff.  It works for me but
there are some trivial changes needed to bogofilter/bogoutil, which
I'll coordinate with David once he's back.  Expect to read more about
this in a week or so.

-- 
| G r e g  L o u i s          | gpg public key: finger     |
|   http://www.bgl.nu/~glouis |   glouis at consultronics.com |
| http://wecanstopspam.org in signatures fights junk email |