Re: Mixed 64-bit system GerH binaries / BillYscripts --> two-sided training? YES!

On Monday, 01 September 2008 at 16:47, Bill Yerazunis wrote:
> 
> What Fidelis said is indeed what I was going to suggest if it were
> not for a bad net connection and a very unhappy 11-month-old.
> 
> But yes, try OSBF for now.  
> 
> Actually, one "good" justification for going to a "train both 
> positive and negative examples" in libcrm114 is that _in the special
> case of just two classes_ .  If you know (ahead of time) that you 
> only want a spam filter (and no advanced email sorting) then 
> you train one statistics file for spam-matching (positive training)
> and that's all.
> 
> Then, classifying with that file will yield a 
> 
>   "Best match to file #0 (spam.css)"
> 
> if it's spam, and the (new)
> 
>   "Best match to file #-1 (none_of_the_above)"
> 
> if it's not spam.
> 
> Advantages:
> 
>   1) most users will need only 1 statistics file (not two)
> 
>   2) if we go with 32-bit features, then the size of the statistics
>      file for Markov / OSB / OSBF will stay the same, but since you
>      only need one, then the overall disk footprint and cache requirement
>      will drop by a factor of two.
> 
>   3) you can still add other file types later.  It just may take a
>      little while to train them in.
> 
>   4) there will no longer be any difference in how you train classifiers;
>      Markovians will train the same as SVMs and neural nets.  No more code
>      dependencies!
> 
>   5) - and you will be able to mix different classifiers in the same 
>      CLASSIFY statement.
> 
> So, Ger, what does this look like in a portable 32/64-bit C:
> 
>    typedef struct mythical_new_bucket
>      { gimme_32_bits  hash;
>        gimme_32_bits  in_count;
>        gimme_32_bits  out_count:
>      } standard_bucketp;
> 
> Do I have to use uint32_t (which can't be printed by printf 
> without casting?  Or is there some other incantation that Just Works.

You could use unsigned int, which is 32bit on virtually every
standard-compliant system. OTOH, uint32_t can be printf'd with %d/%u
(or can you give an example where it can't?).

Regards,
R.

-- 
Fedora http://fedoraproject.org/wiki/User:Rathann
Livna http://rpm.livna.org | MPlayer http://mplayerhq.hu
"Faith manages."
        -- Delenn to Lennier in Babylon 5:"Confessions and Lamentations"

-------------------------------------------------------------------------
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=/