Hi all,
Is it possible to run the tuning process many consecutive times (the next
one use the previous tuning results as the start up moses.ini) to obtain a
better model's weights? (at least when we test on the dev set)?
Tks.

On Mon, Jan 21, 2013 at 11:57 PM, Eleftherios Avramidis <
[email protected]> wrote:

>  Hi, I had written down this, when I worked with EMS (experiment.perl)
> some years ago. I am not sure if it is still applicable, but you can check:
>
>
>    1. If some mert runs have happened (that may meen a couple of days!)
>    you have to let tuning resume the broken process. Open the file
>    steps/TUNING_tune.*num* by using a text editor. Find the line which
>    starts with executing mert-moses.pl . There are two parameters you can
>    add to resume the broken tuning:
>       1. If the error has occurred during decoding, then just add the
>       parameter --continue.
>       2. If a decoding set has successfully happened, and mert crashed
>       during optimization that happens after that, then add the parameter 
> --continue
>       --skip decoder which will save you some time. That's optional,
>       though.
>     2. Save the file and execute the tuning step, (preferably in
>    cluster), separately: nice sh steps/TUNING_tune.*num*
>    3. Once tuning has completed succesfully, delete steps/TUNING_tune.*num
>    *.STDERR and let experiment.perl go on experiment.perl -continue
>    -config steps/config.*num*
>
> best
> Lefteris
>
>
> On 21/01/13 16:42, Arezki Sadoune wrote:
>
>
> Hi
>
> Does anyone know if it is possible to continue or resume a tuning process
> that was accidentally interrupted. ****
>
>
>
> To be more concrete, I lost a MERT tuning process after four weeks of
> running because of an unintended system reboot.
>
> I recovered the weights from the last run file  *( run13.moses.ini)* but
> the translation system I got was not satisfactory.
>
> I think that the translation system will be better if the runing of MERT
> goes to completion
>
> Thank you
>
>
>
> Arezki
>
>
> _______________________________________________
> Moses-support mailing 
> [email protected]http://mailman.mit.edu/mailman/listinfo/moses-support
>
>
>
> --
> MSc. Inf. Eleftherios Avramidis
> DFKI GmbH, Alt-Moabit 91c, 10559 Berlin
> Tel. +49-30 238 95-1806
>
> Fax. +49-30 238 95-1810
>
> -------------------------------------------------------------------------------------------
> Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH
> Firmensitz: Trippstadter Strasse 122, D-67663 Kaiserslautern
>
> Geschaeftsfuehrung:
> Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender)
> Dr. Walter Olthoff
>
> Vorsitzender des Aufsichtsrats:
> Prof. Dr. h.c. Hans A. Aukes
>
> Amtsgericht Kaiserslautern, HRB 2313
> -------------------------------------------------------------------------------------------
>
>
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support
>
>


-- 
*
Best Regards,
C. Hoang

{Mimosa, SMT}@Addict
*
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to