nan in bogofilter stats

[David Relson]

On Tue, 7 Apr 2009 10:37:21 +0930
Stephen Davies wrote:

> Thanks David. I shall set up a similar cron job.
> 
> The zero count does not seem to come from a missing .MSG_COUNT but
> from an explicit entry such as:
> 
> .MSG_COUNT 276248 0 20090322
> 
> How can this come about?
> 
> Cheers,
> Stephen

Off-hand ???  I dunno.

.MSG_COUNT is bogofilter's special token for keeping track of how many
good and bad messages have been registered into the wordlist.  As a
quick rule of thumb, it should have a good count higher than _any_
other token's good count and a bad counter higher ...  

In normal usage as more and more messages are registered with
bogofilter the counts just keep increasing.  The exception is that
bogofilter's -N and -S flags cause a decrease.  

A scenario that would give a zero "good" count is: (1) register just a
few ham messages and (2) unregister that same number of ham messages.

If one registers large quantities of the incoming ham and spam,
then .MSG_COUNT won't ever have a ham or spam count that's even close
to zero.

Some interesting question are:  How do you use bogofilter?  Do
you ever register ham?  Do you just register spam?  Do you use the "-N"
flag?

For what it's worth, I've been using the "-u" (auto-update) flag for 6
or so years.  My mail server's .MSG_COUNT token has spam and ham counts
of 1,624,551 and 181,079.  The 10::1 ratio is higher than we
recommended (6 or 7 years ago) but gives good results (with approx
99.9% of the spam getting caught).  Of course, "-u" has a caveat:  one
must be conscientious about correcting errors (false positives and
false negatives).

Anyhow, I've rambled on long enough.

Hope that this is of value to you.

Regards,

David