Hi again.

The way you tuned seems ok. The final moses.ini file that you found in 
mydata/model/mert-work 

should look like the one you have in 
mydata/model

BUT
with the tuned values for the model's weights. Therefore the one in
mert-work is the moses.ini file you should use to perform your
translations. 

More information about the whole tuning process can be seen in 
http://www.statmt.org/moses/?n=FactoredTraining.Tuning

The
best way you can realize if you did the tuning right is comparing the
results using the different moses.ini files (i.e. using the old and new
weight values).

I don't know about the warning you are getting, sorry.

Good luck!

 --
Carlos A. Henríquez Q.
[EMAIL PROTECTED]
[EMAIL PROTECTED]



----- Mensaje original ----
De: "[EMAIL PROTECTED]" <[EMAIL PROTECTED]>
Para: [email protected]
Enviado: sábado, 22 de noviembre, 2008 16:51:24
Asunto: [Moses-support] some questions about tuning

Hi everybody,

I had previously asked a question on this list if my MOSES setup seemed
reasonable to anybody.  Carlos Henriquez replied, suggesting that I
tried tuning, for which thanks!

Now I am trying this tuning process out and have would be much obliged
if somebody could check if I've got the procedure right.

I have split my original training data into training' and dev.  I ran
trained moses on training' which resulted in a directory

  mydata/model
  mydata/model/moses.ini
  mydata/model/some_other_important_files

I then ran the tuning script as
  cd mydata/model
  ${MOSES_SCRIPTS}/training/mert-moses.pl\
    ../dev-source-language\
    ../dev-target-language\
    /usr/local/bin/moses

This gave me
  mydata/model/mert-work
  mydata/model/mert-work/moses.ini
  mydata/model/merk-work/some_other_important_files

I intend to run moses on the test data as follows
  cd mydata
  moses -f model/mert-work/moses.ini < test-source-language

Does that sound right?

Also, I get this warning

> Use of uninitialized value within @WORD_FACTOR in concatenation (.) or
> string at ${MOSES_SCRIPTS}/training/filter-model-given-input.pl line
> 132, <INPUT> line 1.

Have I done something wrong?  I'm using an unfactored model, for what
it's worth.  Interestingly, if I run the script a second time, I do not
get this complaint.

Thanks!

-- 
Eric Kow <http://www.nltg.brighton.ac.uk/home/Eric.Kow>
PGP Key ID: 08AC04F9



      

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

Reply via email to