Re: size of variables in hyperspace?

Thomas Michael Hagen <[email protected]> Tue, 9 Jun 2009 13:37:56 +0200
Newsgroups gmane.mail.spam.crm114
Message-ID <[email protected]>
it could seem that crm thinks the .chs-file is empty when it reaches a
certain size. i'm now getting the same error on spo.chs, which
contains sports. that file reaches about 2Gb in size after twenty
minutes of learning.


On Tue, Jun 9, 2009 at 9:59 AM, Thomas Michael Hagen
<[email protected]> wrote:
> i'm playing with hyperspace, since it seems osbf won't converge on any
> sort of learning on my 9 + 2 categories of newspaper articles.
>
> hyperspace seemed very promising until i got this error:
>
> crm114: *UNTRAPPABLE ERROR*
> Better start screaming, guv', since the software's just gone critical:
> assertion 'feature_pos + single_featureblock_size <= features_bufferlen' failed!
> Sorry, but this program is very sick and probably should be killed off.
> This happened at line 416 of file
> /sandisk/prosjekt/aviskorpus/emneklassifisering/thomas/hyperspace/learn-or-compare.crm:
>    learn <hyperspace> [:my-args:] (unc.chs)
> (runtime system location: crm_vector_tokenize.c(753) in routine:
> crm_vector_tokenize)
> [st08764@gandalf hyperspace]$
>
>
> i'm not sure what this means, but i'm guessing it has something to do
> with the size of the input variable?
>
> i start the program with:
>
> /usr/local/crm114/bin/crm114 -w 152500000 learn-or-compare.crm dag-filer
>
> where dag-filer is a list of files containing anything from 0 to
> several hundred articles.
>
> the 'unc.chs' file referred to in the error message is the 'uncertain'
> category, which means a classification attempt failed.
>
> the 'learn-or-compare.crm' program learns a number of articles, then
> tries to classify, and compares its results to a gold standard.
>
> so, what sizes should i change to get rid of this problem?
>

------------------------------------------------------------------------------
Crystal Reports - New Free Runtime and 30 Day Trial
Check out the new simplified licensing option that enables unlimited
royalty-free distribution of the report engine for externally facing 
server and web deployment.
http://p.sf.net/sfu/businessobjects