article on blocking by subnets - Justification

[David Relson]

At 06:42 PM 12/5/02, Barry Gould wrote:

>At 12:06 PM 12/5/2002, you wrote:
>>Notes: Be sure to use new wordlists for each run.  The newest cvs 
>>versions of bogofilter allow "block_on_subnet=Yes" to be put into the 
>>config file, which makes testing easier.
>
>This feature doesn't seem to work from the config file:
>
>1277 > ./bogofilter
>/etc/bogofilter.cf:28:  Error - unknown parameter in 'block_on_subnets=no'
>
>1280 > ./bogofilter
>/etc/bogofilter.cf:28:  Error - unknown parameter in 'block_on_subnets=yes'
>
>1278 > ./bogofilter -V
>bogofilter version 0.9.1.cvs.20021205 Copyright (C) 2002 Eric S. Raymond
>
>Also, I just refreshed my CVS checkout, but it still says 0.9.1

I just updated configure.in to correct the version string (now 0.9.1.2) and 
config.c (to include block_on_subnets).  Obviously, it didn't get released 
like I thought it had.  Sorry 'bout that.

>>First, take a month's messages and separate spam from ham.
>>
>>Phase 1:  run script contrib/randomtrain.  Afterwards display MSG_COUNT 
>>from spamlist.db and goodlist.db to determine how many messages were 
>>mis-classified, hence trained on.
>
>With about 5MB of email in each file (spam and ham), the variation due to 
>randomtrains randomness is quite large:
>
>first run:
>                        spam   good
>.MSG_COUNT               77     24
>
>retry first run (deleted db's first):
>                        spam   good
>.MSG_COUNT               62     19
>
>That may make it difficult to use for testing for improvements.

Have you modified randomtrain so that it reuses the shuffle file?  I'm 
about to run some tests of my own, and could probably make the changes if 
you can't