cannot filter virus letters

[David Relson]

On Tue, 10 Feb 2009 13:27:47 +0300
Dmitry wrote:

.....[snip].... 
 
> Hello David,
> 
> The best results I get with the values:
> 
> min_dev=0.1
> robx=0.8
> robs=1.0
> ham_cutoff = 0.45
> spam_cutoff= 0.82
> 
> Now everything seems to be OK. No more virus letters. But there is
> another problem letter sent to mail mailbox that makes me unhappy.
> When I pass it to `bogofilter -s` it becomes more hammy! 
> 
> This is the output of a series "bogofilter -s ; bogofilter -t"
> commands:
> 
> U 0.517247
> U 0.513562
> U 0.510004
> U 0.507321
> 
> What a strange result! It is the opposite of what I expect. The
> content of this letter is commercial spam with all words concatenated
> without spaces. Unfortunately, I can't quote this letter here because
> of non-latin charset. When I switch back to default bogofilter.cf
> with default values, the spammicity of this letter stay always at "U
> 0.500000". Exhaustive training  does not change anything. What can be
> done in such situation?

Indeed the result is strange.  Have you tried running bogofilter with
"-vvv" flags?  That will list the tokens parsed by bogofilter along
with their spam scores.  The FAQ describes the use of "-v" flags and
the output generated.

A message can be zipped and attached to avoid latin/non-latin issues.
 
> Another small problem --  I get errors when I define custom wordlists
> like this:
> wordlist i,ignore,~/.bogofilter/ignorelist.db,1
> wordlist r,wordlist,~/.bogofilter/wordlist.db,2

...[snip]...

> It works If I define full path to wordlists:
> wordlist i,ignore,/home/user/.bogofilter/ignorelist.db,1
> wordlist r,ignore,/home/user/.bogofilter/wordlist.db,2
> 
> Is this expected behaviour? Am I missing something? Should I define
> $HOME variable before running bogofilter command from the console?

Yes.  Bogofilter is a program and can only use the information you
provide for it.