Re: [Fwd: Re: crm114 on 64 bit SuSE]

Thomas Spahni <[email protected]>
Newsgroups gmane.mail.spam.crm114
Message-ID <[email protected]>
On Thu, 14 Aug 2008, Gerrit E.G. Hobbelt wrote:

> !@#$% - hurry means forgot to check the to address / reply-all. This copy is
> FYI.
> 
> -------- Original Message --------
> Subject: 	Re: [Crm114-general] crm114 on 64 bit SuSE
> Date: 	Thu, 14 Aug 2008 14:57:20 +0200
> From: 	Gerrit E.G. Hobbelt <[email protected]>
> Reply-To: 	[email protected]
> Organization: 	Bermuda Holding
> To: 	Thomas Spahni <[email protected]>
> References:
> <561cc61d0808110943i55b62399g478d1299dd24134e-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org><Pine.LNX.4.63.0808120941410.1940-B2ynjq0xCDP9s9LAhoKJkg@public.gmane.org>
> <20080812084035.GA12011@localhost> (sfid-20080812_104055_952859_68B94AF5)
> <Pine.LNX.4.63.0808131819270.9907-B2ynjq0xCDP9s9LAhoKJkg@public.gmane.org>
> 
> Thomas Spahni wrote:
> > > > Any ideas why count of messages learned goes up by 2 after learning one
> > > > message? I never observed this on 32bit systems.
> > > 
> > > which classifier? all? - I suspect a raw bytes count's biting us ...
> >
> > I observed this with 'osb unique microgroom' (default).
> >
> >   
> 
> Hmmm... you're using mailtrainer.crm I suppose? (you do if you use
> mailreaver.crm; not entirely sure but I think you get to use it as well when
> using mailfilter.crm -- this is off the top of my bald spot though)

Correct. Using mailreaver.crm which in turn calls mailtrainer.crm for 
learning.

> In that case (and here Ger starts to talk gibberish like the Delphi 
> Oracle) if _may_ be due to your mailfilter.cf settings which may permit 
> multiple rounds of training on a single message _when_ a training round 
> does not cause the subsequent classify to produce a sufficiently large 
> pR value (training threshold).

The Delphi Oracle may be right. On one occasion I observed that learning 3 
messages resulted in the counter advanced by 5, i.e. 2+2+1

> Don't know the names of those two config parameters in there by head, 
> but you should be looking for a threshold that's probably still at 10.0 
> (while classification thresholds are now two, both I believe are at 5.0 
> these days) and at least one more parameter which permits 
> mailtrainer.crm to loop the train+recheck classification in there. If 
> you need help with this, please kick me after the weekend again; bit 
> busy with an (unrelated) deadline I have to make by the end of this week 
> and it's a bit of a crunch there.

I've set the thresholds to +/- 5.0 now but it behaves still the same.

Thomas

-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.