Hieu Hoang
4894df6494
merge
2013-04-29 19:04:00 +01:00
Hieu Hoang
ce95c117f6
merge
2013-04-29 18:46:48 +01:00
Lane Schwartz
ef3d30146a
Re-enabled commented out code.
...
Hopefully this commit will now compile on all versions of Boost.
2013-04-29 12:24:44 -04:00
Lane Schwartz
62e1bf05d6
Added slightly more unit test coverage for backward LM.
2013-04-29 09:18:46 -04:00
Lane Schwartz
afa270a2e6
Initial unit test for BackwardLanguageModel Evaluate method
2013-04-29 09:18:46 -04:00
Lane Schwartz
e5de2e5917
Checking in changes.
...
This currently doesn't compile.
Need to debug why.
2013-04-29 09:18:46 -04:00
Lane Schwartz
4a44c4682b
Work on BackwardLM unit testing
2013-04-29 09:18:46 -04:00
Lane Schwartz
aacc936204
Adding unit tests to Backward LM
2013-04-29 09:18:46 -04:00
Lane Schwartz
95e3a37469
Added CalcScore to BackwardLanguageModel
2013-04-29 09:18:45 -04:00
Lane Schwartz
d340218fd4
BackwardLM now can calculate its empty state
2013-04-29 09:18:45 -04:00
Lane Schwartz
8459a86137
Refactoring code from Ken.cpp into Ken.h
...
in preparation for implemenation of backward language model
from Xiong et al (2011)
2013-04-29 09:18:45 -04:00
Hieu Hoang
f5d3f16fab
rescale LM score for chart decoding by KenLM. All regression tests passes
2013-04-29 13:01:26 +01:00
Kenneth Heafield
972001e345
Compiler error
2013-04-27 01:02:52 +02:00
Hieu Hoang
aa9689d12b
rescale LM score for chart decoding
2013-04-26 19:39:29 +01:00
Hieu Hoang
8edbfebbf1
word penalty for XML options
2013-04-26 17:49:11 +01:00
Hieu Hoang
83494418e2
delete PhraseDictionaryMemory. Fix word penalty in compact phrase table
2013-04-26 15:55:32 +01:00
Hieu Hoang
69453f0a8b
don't convert LM score from 1 log to another for chart decoding
2013-04-26 14:27:44 +01:00
Hieu Hoang
b448c6285a
make scfg work with factored input. Strip factors from source that are not part of phrase-table input factors
2013-04-26 12:20:49 +01:00
Kenneth Heafield
b836753ec2
Merge branch 'master' of github.com:moses-smt/mosesdecoder
2013-04-25 22:52:06 +02:00
Kenneth Heafield
38b29e2c79
Merge branch 'master' of github.com:moses-smt/mosesdecoder
2013-04-25 22:51:26 +02:00
Kenneth Heafield
2437ed441d
Accept cmph with lib64 too
2013-04-25 21:50:05 +01:00
Kenneth Heafield
f1d3663810
Back FactorCollection with a memory pool. Less memory for large vocabularies.
2013-04-25 19:42:30 +01:00
Hieu Hoang
75c85acd5e
delete PhraseDictionaryMemory. Use PhraseDictionarySCFG
2013-04-25 18:24:02 +01:00
Hieu Hoang
4b4aba3d81
log probabilities. /Abby
2013-04-25 16:54:45 +01:00
Rico Sennrich
4e87a012d0
fix two bugs with relax-parse:
...
- size of sentence was not calculated correctly
(instead, number of positions at which a subtree starts was used)
- code entered an infinitive loop sometimes; added break condition
2013-04-25 17:27:50 +02:00
Hieu Hoang
b53b93f09f
gt rid of word penalty of 0.434. Adjust weight accordingly
2013-04-24 19:23:14 +01:00
Hieu Hoang
6da2cf7663
eclipse
2013-04-24 14:19:16 +01:00
Hieu Hoang
2d265344d1
delete m_currScoreBreakdown
2013-04-24 13:44:11 +01:00
Hieu Hoang
d9b1784373
delete different calc functions
2013-04-24 13:23:22 +01:00
Hieu Hoang
c84605a241
delete different calc functions
2013-04-24 13:13:50 +01:00
Hieu Hoang
16cbf692d3
delete different calc functions
2013-04-24 12:10:56 +01:00
Kenneth Heafield
8a1e944bb4
Error messages
2013-04-23 08:36:24 +01:00
Hieu Hoang
3248a8e4f4
initialise m_currScoreBreakdown with score from transOpt
2013-04-22 21:25:36 +01:00
Hieu Hoang
5126aa8e30
word penalty is a cached stateless feature
2013-04-22 21:17:52 +01:00
Hieu Hoang
bf057dc46b
get rid of ResetScore() from Hypothesis and TargetPhrase class
2013-04-22 20:12:43 +01:00
Hieu Hoang
5655a6cf1b
get rid of tempScoreBreakdown. Not sure what it was for
2013-04-22 18:38:42 +01:00
Hieu Hoang
3bfcdd0077
minor changes for debuggablility
2013-04-22 13:03:07 +01:00
Rico Sennrich
908c006e32
online combination of multiple phrase tables
...
- creates a virtual phrase table at decoding time based on a vector of component models and a combination algorithm
- linear interpolation or instance weighting
- two possible component model types supported so far: 0 (in-memory) or 12 (compact)
- weights can be set in config, and overriden on a sentence-level through mosesserver API
- online optimization (perplexity minimization) using dlib and xmlrpc-c call
2013-04-22 13:21:59 +02:00
Christian Buck
477f913585
mosesserver now supports nbest lists
2013-04-21 00:34:59 +01:00
Christian Buck
b362791169
fixed warnings
2013-04-21 00:34:25 +01:00
Christian Buck
046e031723
Merge branch 'master' of git://github.com/moses-smt/mosesdecoder
2013-04-20 20:33:16 +01:00
Hieu Hoang
b1da4dbe0e
merged
2013-04-19 15:03:34 +01:00
Barry Haddow
7677832655
Fix markdown
2013-04-19 09:00:24 +01:00
Barry Haddow
71295e2708
correct markdown
2013-04-19 08:45:03 +01:00
Barry Haddow
a8fe122e4b
Merge branch 'master' of github.com:moses-smt/mosesdecoder
2013-04-19 08:35:01 +01:00
Barry Haddow
e53faa174c
Complete README
2013-04-19 08:34:43 +01:00
Hieu Hoang
9749ca5fd9
Fixed blank lines when using -print-alignment-info
2013-04-18 16:38:52 +01:00
Barry Haddow
2ba633c5f8
Create small binarised models for testing
2013-04-18 07:51:06 +01:00
Barry Haddow
13d21428ed
stub out README
2013-04-17 21:57:09 +01:00
Barry Haddow
e6cabf802d
remove absolute path default
2013-04-17 21:35:32 +01:00