> On 02/04/2013 20:42, Rob Voigt wrote:
> > Hi all,
> >
> > I'm trying to use confusion networks with Moses - I can decode with them 
successfully, manually inputting
> an insertion weight (-weight-i), but am running into trouble doing MERT on 
a model that uses confusion networks.
> >
> > Running this command, with weight-i specified in the moses.ini file:
> > nohup nice ~/mosesdecoder/scripts/training/mert-moses.pl tune.zh tune.en
> ~/mosesdecoder/bin/moses moses.ini --decoder-flags="-threads 8 -inputtype 
1 -drop-unknown"
> --inputtype 1 --mertdir ~/mosesdecoder/bin/ >& CN_mert.out
> >
> > I get through one iteration of training, but then get the following 
error during decoding on the second
> iteration of MERT:
> > ERROR:You specified 6 input weights (weight-i), but you specified 1 link 
parameters (link-param-count)!
> >
> > Looking at the run1.moses.ini file, it looks like five additional 
weight-i values have been added.
> >

[...]

> >
> > How does one use MERT if the input is a confusion network? Is it 
possible?
> >
> > Thanks very much for your time!
> >
> > -Rob

Hi Rob/Hieu,

the problem is with the file features.list that mert-moses.pl creates 
(unless the file already exists) by asking moses to dump a list of features. 
Right now, the list looks something like this:

--- features.list (=>crash) --------------------------------
Distortion d 0.300000
LexicalReordering_wbe-msd-bidirectional-fe-allff d 0.300000
LexicalReordering_wbe-msd-bidirectional-fe-allff d 0.300000
LexicalReordering_wbe-msd-bidirectional-fe-allff d 0.300000
LexicalReordering_wbe-msd-bidirectional-fe-allff d 0.300000
LexicalReordering_wbe-msd-bidirectional-fe-allff d 0.300000
LexicalReordering_wbe-msd-bidirectional-fe-allff d 0.300000
LM lm 0.500000
WordPenalty w -1.000000
PhraseModel I 0.100000
PhraseModel I 0.200000
PhraseModel I 0.200000
PhraseModel I 0.200000
PhraseModel I 0.200000
PhraseModel I 0.200000
-------------------------------------------------------------

The last five lines of the file should be 

PhraseModel tm 0.200000
PhraseModel tm 0.200000
PhraseModel tm 0.200000
PhraseModel tm 0.200000
PhraseModel tm 0.200000

Then at least tuning runs. I have no idea if it runs as intended, though.

This is probably a bug in moses' feature reporting functionality.
You can work around the problem by manually editing the file features.list 
in the tuning directory (usually tuning/tmp.<step> when you use ems).

- Uli



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

Reply via email to