Commit Graph

470 Commits

Author SHA1 Message Date
Hieu Hoang
8e2c4a89c0 Merge ../mosesdecoder into hieu 2014-06-04 12:53:06 +01:00
Ulrich Germann
9c27dc08e9 Merge branch 'master' of https://github.com/moses-smt/mosesdecoder
Conflicts:
	moses/FF/ConstrainedDecoding.cpp
2014-06-03 18:55:53 +01:00
Ulrich Germann
9cda39f802 Major bug fix in setting ug_bitext::jstats.my_cnt2. 2014-06-03 15:48:27 +01:00
Ulrich Germann
8c9a26a3be Added phrase feature function for log counts. 2014-06-03 15:45:29 +01:00
Ulrich Germann
0fc80660f3 TSA_tree_iterator::approxOccurrenceCount() now returns exact count for small values. 2014-06-03 15:44:20 +01:00
Ulrich Germann
a248349da3 Added more phrase table features and configuration options to Mmsapt. 2014-06-03 15:41:59 +01:00
Ulrich Germann
ce435d22bf Mmsapt now adds the phrase penalty only if num_features specified
on the respective line in moses.ini is odd.
2014-06-01 18:23:00 +01:00
Ulrich Germann
249c40ec74 Eliminated factor delimiter from the parameters of
Phrase.CreateFromString() and Sentence.CreateFromString(),
as it was never used in those functions anyway ---
Word.CreateFromString() retrieves the factor delimiter
from StaticData directly.
2014-05-31 14:33:31 +01:00
Ulrich Germann
c452875343 Mmsapt/ug_bitext:
- Sampling now uses a static seed for consistency between runs.
- (In order to have consistency guarantees, specify workers=1 in the
- Mmsapt line in moses.ini; otherwise the randomness of thread scheduling may
  cause inconsistencies between runs.
- Denominator for fwd phrase probabilities can now be chosen via an
  external argument.
2014-05-31 14:29:44 +01:00
Ulrich Germann
380b2da2d0 Bug fix in and in connection with Mmsapt::combine_pstats(). 2014-05-19 21:34:58 +01:00
Hieu Hoang
35c812f5df Merge ../mosesdecoder into hieu 2014-05-13 18:24:17 +01:00
Hieu Hoang
bf7c033271 backoff for default GetTargetPhraseCollectionBatch(), used by binary pt and compact pt 2014-05-12 15:50:32 +01:00
Hieu Hoang
2830ed9968 use Nadir's backoff code 2014-05-12 15:40:18 +01:00
Hieu Hoang
1d401817a5 backoff fix for memory and on-disk phrase-tables 2014-05-12 15:16:11 +01:00
Hieu Hoang
4e977f0f1f Merge ../mosesdecoder into hieu 2014-05-07 12:21:49 +01:00
Hieu Hoang
b436570ab0 delete unused files 2014-05-07 12:21:19 +01:00
Hieu Hoang
71242a3578 Merge ../mosesdecoder into hieu 2014-05-07 12:11:24 +01:00
Hieu Hoang
2beed771d2 minor error. Segfault if phrase length is bigger than max-phrase-length 2014-05-01 16:47:17 +01:00
Hieu Hoang
17c10d45f3 merge 2014-05-01 14:52:38 +01:00
Ulrich Germann
7c145d045b Merge branch 'master' into dynamic-phrase-tables
Conflicts:
	contrib/server/Jamfile
	contrib/server/mosesserver.cpp
2014-04-28 10:00:07 +01:00
Hieu Hoang
da6ade7d94 get-by-line-number.perl 2014-04-24 17:48:12 +01:00
Ulrich Germann
e5b739946f Merge branch 'master' of https://github.com/moses-smt/mosesdecoder 2014-04-23 18:29:39 +01:00
Hieu Hoang
00505ba048 minor leak 2014-04-23 11:36:25 +01:00
Ulrich Germann
24545ba1de Changes due to new pure virtual function in PhraseTable. 2014-04-11 04:28:34 +01:00
Ulrich Germann
21f141ce0d Merge branch 'master' into dynamic-phrase-tables
Conflicts:
	moses/InputPath.cpp
	moses/Word.cpp
2014-04-08 14:58:29 +01:00
Hieu Hoang
b98dac2971 merge 2014-04-06 19:07:13 +01:00
Rico Sennrich
1f435340f0 faster pruning in chart decoding 2014-03-26 11:23:59 +00:00
Ulrich Germann
f63606a92b 1. Renamed PhraseDictionary member function release() to Release().
2. Added option to check phrase prefixes for early stopping of
   InputPath generation when translating confusion networks.
2014-03-25 01:49:24 +00:00
Ulrich Germann
80b031769d (1) Bug fixes related to locking. (2) pstats cache disabled when translating confusion networks. 2014-03-25 01:48:49 +00:00
Ulrich Germann
84299006da Added another constructor to TSA_tree_iterator. 2014-03-25 01:43:00 +00:00
Ulrich Germann
f5d969ab37 Added missing library for custom-pt.cc 2014-03-25 01:42:06 +00:00
Ulrich Germann
728ceb4e4c Added missing #include. 2014-03-24 14:51:51 +00:00
Ulrich Germann
14e52c9b94 Errors occurring when accessing data files now throw exceptions. 2014-03-24 13:39:15 +00:00
Ulrich Germann
ed825cd234 Bug fix: dependency .../UG/generic//generic was missing for custom-pt 2014-03-24 13:33:46 +00:00
Ulrich Germann
20aa9dd189 Adapted to configurable map type in pstats. 2014-03-24 13:32:41 +00:00
Ulrich Germann
794defa238 Numerous bug fixes; implemented caching and improved life cycle
management for TargetPhraseCollections.
2014-03-24 13:30:22 +00:00
Ulrich Germann
7fd87943ea Reinstated pstats caching. 2014-03-24 13:29:05 +00:00
germann
2063da274d Made member Tokenindex.ridx mutable. 2014-03-24 12:34:59 +00:00
germann
5ec4436b07 Initial check-in. 2014-03-24 12:33:17 +00:00
germann
c12c25770a Added virtual member function release(TargetPhraseCollection*) to
PhraseDictionary, which gets called by the destructor of InputPath to
release TargetPhraseCollections that aren't needed any more.
2014-03-24 12:29:46 +00:00
Hieu Hoang
e5a6d63cf2 Merge branch 'master' of github.com:moses-smt/mosesdecoder into hieu 2014-03-21 21:41:04 +00:00
Ulrich Germann
1f32bd0d08 Fixed header inclusions. 2014-03-21 16:49:44 +00:00
Rico Sennrich
fb16df8c00 typo in last commit 2014-03-21 11:59:41 +00:00
Rico Sennrich
45630a5851 various optimizations to make CYK+ parser several times faster and eat less memory.
speed-up of decoding depends on how much time is spent in parser:
10-50% speed-up for string-to-tree systems observed (more on long sentences and with high max-chart-span).

if you only use hiero or string-to-tree models (but none with source syntax), use compile-option --unlabelled-source for (small) efficiency gains.
2014-03-21 11:12:24 +00:00
Ulrich Germann
4685aeadd2 Bug fix. 2014-03-21 11:11:24 +00:00
Ulrich Germann
d5c96eff35 Minor edit of a comment. 2014-03-21 11:08:37 +00:00
Ulrich Germann
f030986f51 Bug fixes. Added some sanity checks. 2014-03-21 11:07:17 +00:00
Ulrich Germann
43e6c68586 Changed default method for PhraseScoreFwd. Added minor debugging message. 2014-03-19 20:54:49 +00:00
Ulrich Germann
7e344d1c53 1. Added variant of CreateRuleLookupManager variant as required by the new
pure virtual function in PhraseTable.h (from the master branch).

2. Bug fix in pool_stats.
2014-03-19 20:43:47 +00:00
Ulrich Germann
c91fb5cc84 Added mmlex-lookup utility program. 2014-03-19 20:41:29 +00:00