Ulrich Germann
dcffbb5f4d
Made LRModel::ReorderingType an enumerated type.
2015-03-16 00:24:11 +00:00
Phil Williams
e8a7163f0d
Add SyntaxInputWeight feature function
...
Currently only used for forest input.
2015-03-10 11:07:04 +00:00
Ulrich Germann
c1d2313a66
Moved caching of lexical reordering scores from class TranslationOption to class TargetPhrase.
...
This was done so that phrase tables can add this information (if available) as extra annotation
to TargetPhrases, in preparation of providing lexical reordering models with sampling phrase tables.
2015-03-09 00:30:01 +00:00
Ales Tamchyna
6058bfc61b
bugfix, BP only applies when candidate is shorter than ref :)
2015-03-04 18:02:50 +01:00
Ales Tamchyna
ff1cae919b
minor refactoring in VW feature
2015-03-04 17:40:05 +01:00
Ales Tamchyna
34649b74d3
implement BLEU2+1 as training loss for VW
2015-03-04 17:10:59 +01:00
Ales Tamchyna
7601618477
some refactoring of VW, towards customizable loss calculation in training
2015-03-04 14:26:26 +01:00
Matthias Huck
2543c78b71
RulePairUnlexicalizedSource feature function
2015-03-02 14:32:18 +00:00
Hieu Hoang
0829e5196d
path arg /Cheng Yong
2015-02-28 13:59:36 +00:00
Matthias Huck
0d4678577b
floor value, glue rules, some cleanup
2015-02-27 02:39:43 +00:00
Matthias Huck
0afc261251
SoftSourceSyntacticConstraintsFeature: Sparse label pair scores
2015-02-26 20:27:02 +00:00
Matthias Huck
342abcf8fc
Model1Feature: score caching for improved efficiency
2015-02-25 23:13:15 +00:00
Matthias Huck
0a6adcde3a
Model1Feature: special UNK token in MGIZA vocabularies
2015-02-25 01:27:08 +00:00
Matthias Huck
39c1ef52dc
Model1Feature: unnecessary includes & special UNK token in MGIZA vocabularies
2015-02-25 00:52:15 +00:00
Matthias Huck
3c8d48f8af
Model1Feature: a simple IBM Model 1 scorer,
...
source-to-target with global source-sentence context
2015-02-25 00:11:31 +00:00
Matthias Huck
372d14a9e8
missing condition
2015-02-20 21:46:38 +00:00
Matthias Huck
f317a2cebd
PhraseOrientationFeature
...
better efficiency via precomputation of non-terminal orientation classes,
plus a bugfix (weights vector was not read correctly)
2015-02-20 20:53:50 +00:00
Hieu Hoang
32de075022
beautify
2015-02-19 12:27:23 +00:00
Hieu Hoang
add14db330
use VERBOSE instead of cerr
2015-02-18 21:56:35 +00:00
Matthias Huck
6028c7cf9c
tuneable-components config parameter for feature functions
2015-02-16 18:39:25 +00:00
Ales Tamchyna
a46daa1938
target (internal) bigrams for VW
2015-02-16 11:36:16 +01:00
Ales Tamchyna
b05fc4c9bc
properly register bigram VW feature
2015-02-16 11:30:36 +01:00
Ales Tamchyna
a9926e7f49
source bigram feature for VW
2015-02-16 10:59:55 +01:00
Ulrich Germann
064b01c054
Code reorganization and cleanup around Lexical Reordering Round 2.
...
Hope it compiles this time.
2015-02-15 16:34:54 +00:00
Ulrich Germann
ccf44f39fb
Code cleanup and reorganization. A few classes have been renamed to shorter names.
2015-02-15 01:45:22 +00:00
Ulrich Germann
e520a3d39e
Code cleanup.
2015-02-14 04:52:06 +00:00
Ulrich Germann
fa3f82a04d
Merge branch 'master' into sampled-lexicalized-reordering
2015-02-13 02:22:37 +00:00
Phil Williams
33df7fd2b4
Rename StaticData::IsChart() to IsSyntax() and add new SearchAlgorithm values
2015-02-12 13:03:26 +00:00
Ulrich Germann
4bef4a8a51
Reformatting to improve code readability.
2015-02-09 23:17:12 +00:00
Ulrich Germann
d3e39c2138
Fixed minor issues that triggered compiler warnings.
2015-02-09 23:16:00 +00:00
Hieu Hoang
ce80e53b30
set scores for lattice path
2015-02-09 11:43:44 +00:00
Hieu Hoang
4145999e33
virtual destructor for KenOSMBase /Jeroen Vermeulen
2015-02-06 15:37:47 +00:00
Hieu Hoang
70e8eb54ce
Using boost for prefix/suffix checks /Jeroen Vermeulen
2015-02-05 16:23:47 +00:00
Ales Tamchyna
fd61d2f810
vw bugfix
2015-02-04 11:33:10 +01:00
Ales Tamchyna
e932e96db7
avoid duplicate generation of source-side features when training VW
2015-02-04 11:07:07 +01:00
Matthias Huck
a2efb9afc7
PhraseOrientationFeature: more sparse scores
2015-01-29 20:23:41 +00:00
Ales Tamchyna
c33e9e81da
vw verbose
2015-01-29 17:40:22 +01:00
Ales Tamchyna
2b0aa56677
VW verbose outputs
2015-01-29 17:33:35 +01:00
Ales Tamchyna
79f7e7bd16
Merge branch 'master' of github.com:moses-smt/mosesdecoder
2015-01-29 16:52:27 +01:00
Ales Tamchyna
bef4795ad9
optionally leave one out in VW training
2015-01-29 16:52:21 +01:00
Matthias Huck
7a7d12b038
PhraseOrientationFeature: can produce some sparse scores
2015-01-28 20:50:24 +00:00
Matthias Huck
91dfd36c8b
use FEATUREVERBOSE
2015-01-28 20:49:04 +00:00
Matthias Huck
9f562e0fd4
SoftSourceSyntacticConstraintsFeature: better config parameter names
2015-01-27 18:15:51 +00:00
Matthias Huck
0a0ea437bb
use pragma once; avoid using cerr directly
2015-01-26 22:12:44 +00:00
Matthias Huck
eb9d8134a7
PhraseOrientationFeature: Heuristic score for boundary non-terminals is basically a lookahead. Compute a lookahead for everything.
...
(+ Refined feature state comparison.)
2015-01-26 21:11:37 +00:00
Matthias Huck
e51714ff7a
a plain dense unaligned word count feature with two scores (source and target unaligned words)
2015-01-26 21:06:12 +00:00
Matthias Huck
c66d6a9b86
using pragma once and VERBOSE in SourceWordDeletionFeature/TargetWordInsertionFeature
2015-01-26 20:45:08 +00:00
Matthias Huck
9987beb453
SoftSourceSyntacticConstraintsFeature: Now for both non-terminals (as before) _and_ terminals.
...
Also added score components based on relative frequency.
(TODO: logprobs right now; are plain probabilities better?)
2015-01-23 18:41:18 +00:00
Hieu Hoang
45ff417244
beautify
2015-01-22 22:41:56 +00:00
Hieu Hoang
ad6f3a8026
option to sort translation options after EvaluateAfterSourceContext
2015-01-22 12:25:02 +01:00