Hi,

quite an old email I am digging out!

On Wed, Oct 5, 2011 at 10:35 PM, Christophe Servan
<[email protected]> wrote:
> Dear All,
> I just put on the trunk the new merge scorer I presented to the MT
> Marathon in Trento.
> This scorer allows to tune with multiple metrics i.e. 2xBLEU+1xTER
> In order to use it, I also the mert-moses-multi.pl that come from the
> mert-moses.pl with a new switch.
> This switch can be use like this for the previous exemple of combination :
> --sc-config="BLEU:2,TER:1"
> Let me know if you have any problems with this script.
>

I have not tried this script but I was just wondering why this is a
separate script. If one doesn't use the --sc-config switch, is it
supposed to act exactly as mert-moses.pl already does? If so, I
believe it should replace mert-moses.pl (and if not, I would suggest
to fix so that it does, unless there is a good reason).

I say this because I was wondering what was the difference between the
2 scripts (plus, the name gave me at first the wrong impression it was
about multi-threading (though opening the script showed me that
mert-moses.pl also has multithread support). This could be avoided in
the future if both scripts were merged. Moreover there may be some
divergence in their feature with passing time, so better doing this
when it is still easy. Factorization of code is better to keep a sane
code. :-)

I also see this option in the multi version, but not the old one:
"lattice-samples=i" => \$___LATTICE_SAMPLES,
I guess this is already a divergence which should not have happened, isn't it?

Jehan

> Best Regards,
>
> Christophe
>
> --
> ========================================================
> Christophe SERVAN - PhD - Postdoc
> Laboratoire Informatique de l'Université du Maine (LIUM)
> Institut Claude Chappe - 72085 Le Mans Cedex 9, France
> Tel : 02 43 83 38 52
> Cel : 06 15 73 09 61
> web : http://www-lium.univ-lemans.fr/~servan
> Mail : [email protected]
> ========================================================
>
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support

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

Reply via email to