meaningful or not?

[Greg Louis]

On 20030214 (Fri) at 1629:14 -0500, David Relson wrote:

> I've also run mkdb and runex with the results shown below.  Do they seem 
> reasonable or are they all messed up?  Would you care to hazard an 
> interpretation as to their meaning?
> 
> .MSG_COUNT             4587  17060

> [relson at osage runex.0214.d]$ date ; runex
> Fri Feb 14 16:16:36 EST 2003
> 
> n   0.025 fpos... 4 at cutoff 0.931959, run0...  5, run1...  4
> n   0.050 fpos... 4 at cutoff 0.933136, run0...  5, run1...  4
> n   0.075 fpos... 4 at cutoff 0.979940, run0...  2, run1...  1
> n   0.100 fpos... 4 at cutoff 0.979940, run0...  2, run1...  1
> n   0.125 fpos... 4 at cutoff 0.979940, run0...  2, run1...  1
> 
> y   0.025 fpos... 4 at cutoff 0.946613, run0...  4, run1...  4
> y   0.050 fpos... 4 at cutoff 0.946613, run0...  5, run1...  4
> y   0.075 fpos... 4 at cutoff 0.991569, run0...  0, run1...  0
> y   0.100 fpos... 4 at cutoff 0.991569, run0...  0, run1...  0
> y   0.125 fpos... 4 at cutoff 0.993307, run0...  0, run1...  0

The cutoffs look reasonable, but the fn's are a spam hater's dream :)

Judging by your earlier results, it appears as though you might be
counting only "ternary" fn's, ie spams classed as nonspam but not spams
classed as unsure -- which have to be regarded as fn in a binary sense.

You sent me copies of your versions of mkdb and runex, and they look
fine; I didn't see any problems with either -- but I'm not familiar
with the operations of the config file logic, nor with current
bogofilter in binary mode.  Those being the differences between your
way of testing and mine, I'd naturally look for something there first.
As a first step I'd print the output of the bogofilter call in function
wrapper to make sure classification is happening correctly, with a 1 at
the beginning of every record where the message should have been
classed as spam (spamicity > cutoff), and something other than 1 as the
first character of every record where the message should not have been
classed as spam based on the cutoff.

-- 
| G r e g  L o u i s          | gpg public key:      |
|   http://www.bgl.nu/~glouis |   finger greg at bgl.nu |
| Help free our mailboxes. Include                   |
|        http://wecanstopspam.org in your signature. |