16.2 not as effective

[David Relson]

On Tue, 20 Jan 2004 08:24:52 -0800
Dan Singletary <dvsing at sonicspike.net> wrote:

> Has anyone noticed that bogofilter 16.2 is just not as effective
> against 
>   catching spam as previous versions?  I am not sure when this change 
> happened, but I suspect it occured betwen the 15.x/16.x version jump. 
> With 15.x I would have to manually classify one.. MAYBE two junk mails
> a day, and I currently receive on the order of 100-200 junk mails a
> day which bogofilter catches.  With 16.x I am having to suddenly
> manually classify at least 10 a day.
> 
> Any idea why this would be happening?  I am tempted to go and revert
> to previous versions to figure out exactly where the change occured,
> but I figured I'd ask first to see if anyone knew what was going on.
> 
> -Dan

Hi Dan,

Read RELEASE.NOTES-0.16.  The jump from 0.15.13 (stable) to 0.16.0 just
added "#ifdef ENABLE_DEPRECATED_CODE" statements to turn off a lot of
code.  AFAIK there are no changes that would affect a message's score
(more than a tiny, tiny amount).

As a test, you could install 0.15.13 (renaming the executables to
bogofilter-15 and bogolexer-15) and 0.16 (for bogofilter-16 and
bogolexer-16) and then run some tests.

There have been some parsing changes since 0.16.0, but they're darn near
trivial.  For example, CC: lines are now tagged the same as To: lines. 
If you want to see all the changes, unpack the bogofilter-0.15.13 and
bogofilter-0.16.3 tarballs and run "diff -ur" on the two trees.  (Note: 
you might want to exclude generated files like configure, Makefile,
lexer_v3.c, *.html)

I'd start by saving those manually classified messages in their own
directory.  Then you could run "bogofilter-15 -v -B dir" and
"bogofilter-16 -v -B dir" and compare scores.

Let me know what your tests reveal.

David

P.S. 0.16.3 is the current release.