Hi Guchun,

 

sorry - my mistake - I meant that there probably has gone s.th. wrong
before your preparation step (not alignment step). 

Still your files indicate that there is something wrong with the
previous processes to factorization, pos-tagging ,truecasing and so on.
Maybe the tokenizing or the cleaning didn't work on the English side, so
it didn't give any output? Or, as Barry said, maybe even the corpus that
is input for the whole processing was not found at the beginning? 

 

Von: [email protected] [mailto:[email protected]] Im Auftrag von
Guchun Zhang
Gesendet: Dienstag, 31. Januar 2012 15:14
An: Noubours, Sandra
Betreff: Fwd: [Moses-support] EMS on SGE crashes

 

Many thanks, Sandra and Barry.

 

To Sandra, there is no alignment model generated, which can be related
to Barry's question.

 

To Barry. I checked under the corpus directory. The factored, pos and
truecased English corpora are empty, while all the French corpora are
populated. And all the files are available to any node on the cluster. I
am using NFS.

 

Cheers,

 

Guchun

 

On 31 January 2012 11:34, Barry Haddow <[email protected]>
wrote:

Hi Guchun

It looks as though the prepare data step is unable to find or open your
corpus.

What does the prepare-data script look like? In particular, does the
file
referenced by the -corpus argument exist, and and is it available to the
cluster nodes?

cheers - Barry


On Friday 27 January 2012 17:33:50 Guchun Zhang wrote:
> Hi there,
>
> I am trying to run EMS for factored training on SGE. The config is
simple
> -- use generic-parallelizer.perl and set job number to 2. No other
settings
> for SGE.
>
> At the training_prepare-data step, it crashed. The error given from
the
> stderr file is
>
>
PATH=/usr/lib/lightdm/lightdm:/usr/local/sbin:/usr/local/bin:/usr/sbin:/
usr
> /bin:/sbin:/bin:/usr/games: Command not found.
> Using SCRIPTS_ROOTDIR: .....
> ......
> ......
> (1.0.5) ......
> Use of uninitialized value $firstline in substitution (s///) at
> /home/guchun/Work/moses/moses-scripts/training/train-model.perl line
646.
> Use of uninitialized value $firstline in substitution (s///) at
> /home/guchun/Work/moses/moses-scripts/training/train-model.perl line
646.
> ......
> ......
> (1.3) ......
> Use of uninitialized value $txt in scalar chomp at
> /home/guchun/Work/moses/moses-scripts/training/train-model.perl line
807.
> Use of uninitialized value $txt in split at
> /home/guchun/Work/moses/moses-scripts/training/train-model.perl line
810.
> .
> .
> .
> .
> .
> (1.3) numberizing corpus.......
>
> I checked $PATH on the execution nodes. The value on those nodes is
without
> the lightdm path entry. Does this $PATH difference cause the problem?
Or is
> it something else more profound?
>
> Any advice will be much appreciated.
>
> Have a nice weekend.
>
> Many thanks,
>
> Guchun
>

--
Barry Haddow
University of Edinburgh
+44 (0) 131 651 3173 <tel:%2B44%20%280%29%20131%20651%203173> 

--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.



 

 

_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to