Hi,
we haven't used this in a while, but you seem to be on the right track.
For lattice input, it is not possible to filter the phrase table -
partly because
listing all acceptable input phrases is computationally daunting.
The solution is hence to binarize the entire unfiltered phrase table, which
is fine, may just take a while and take a good amount of disk space.
This is the idea behind the binarize-config step in EMS. You have to check
this closer, but I recall that you simply have to set binarize-all to what
usually ttable-binarizer is set to, i.e.,
$moses-src-dir/misc/processPhraseTable
-phi
On Wed, Mar 30, 2011 at 2:14 PM, Sriram V sriram.venkatapa...@gmail.com wrote:
Hi,
I have been trying to do Mert on lattices. I set the appropriate flags,
--
tuning-settings = -mertdir $moses-src-dir/mert --no-filter-phrase-table
decoder-settings = -inputtype 2 -weight-i 0
--
But the tuning step crashes as it does not find a binary phrase table. There
is an appropriate binarization step indicated in
$SCRIPTS_ROOTDIR/ems/experiment.meta
which is :
--
binarize-config
in: config
out: bin-config
pass-unless: binarize-all
rerun-on-change: config
default-name: model/moses.bin.ini
template: $binarize-all $ttable-binarizer $rtable-binarizer OUT IN
---
But, ems doesn't make use of it, most definitely because $binarize-all is not
defined. Any idea about the script that's being referred by $binarize-all ?
As an alternative, I tried binarizing seperately after the crash, changed the
configuration file and then asked EMS to use the new configuration file
instead
of training again but that seemed to crash also for some reason. I am trying
to figure out why.
I saw a similar question posted earlier last year on moses-support but that
didn't help me much. Any suggestions on this ?
Regards,
Sriram
___
Moses-support mailing list
Moses-support@mit.edu
http://mailman.mit.edu/mailman/listinfo/moses-support
___
Moses-support mailing list
Moses-support@mit.edu
http://mailman.mit.edu/mailman/listinfo/moses-support