Rico Sennrich
6aac7ded9a
EMS: more flexible way to concatenate LM training data.
...
the implementation allows the user to specify which corpora to combine,
and to have multiple LMs on the same data.
2015-05-20 17:20:02 +01:00
Rico Sennrich
8ca6764c7d
ems: allow LMs with user-specified training commands and moses.ini config entries
...
intended for neural LMs, syntactic LMs, and the like. currently doesn't play nice with INTERPOLATED-LM.
2015-05-18 19:07:37 +01:00
Rico Sennrich
fb06a2325e
fix broken ems with interpolated lm disabled
2015-05-18 17:26:09 +01:00
Rico Sennrich
f85dd85f6b
ignore-unless magic
2015-05-18 16:17:33 +01:00
Rico Sennrich
59376f500b
still confused about pass-unless vs. ignore-unless
2015-05-18 14:40:56 +01:00
Rico Sennrich
45a97f9016
EMS: disable concatenated LM by default
2015-05-18 14:10:29 +01:00
Rico Sennrich
27fd45d088
ems: training LM on concatenation of all LM training corpora
2015-05-18 12:18:49 +01:00
Jeroen Vermeulen
e2a632a2b8
JavaScript lint.
2015-05-17 21:36:07 +07:00
Jeroen Vermeulen
5d0bbb6a45
Fix some JavaScript lint. Still a lot left.
2015-05-17 21:24:04 +07:00
Jeroen Vermeulen
a25193cc5d
Fix a lot of lint, mostly trailing whitespace.
...
This is lint reported by the new lint-checking functionality in beautify.py.
(We can change to a different lint checker if we have a better one, but it
would probably still flag these same problems.)
Lint checking can help a lot, but only if we get the lint under control.
2015-05-17 20:04:04 +07:00
Jeroen Vermeulen
61162dd242
Fix more Python lint.
...
Most of the complaints fixed here were from Pocketlint, but many were also
from Syntastic the vim plugin.
2015-05-16 17:26:56 +07:00
Hieu Hoang
abfc0671a3
osm tweaks and morfessor wrapper
2015-05-12 20:19:39 +04:00
Hieu Hoang
8bb18b9ff0
add no-splitter-training argument. Splitter to be used by mada
2015-05-11 15:26:50 +04:00
Barry Haddow
85c1af4d72
Merge branch 'master' of github.com:moses-smt/mosesdecoder
2015-05-08 09:16:55 +01:00
Barry Haddow
f403f5e478
mmsapt doesn't require feature weights on first tuning iteration
2015-05-08 09:16:51 +01:00
Hieu Hoang
2acb590394
output bleu for multi-bleu hack
2015-05-05 17:54:35 +04:00
Hieu Hoang
d006c6ef8c
don't output remaining args twice
2015-05-05 12:15:08 +04:00
Hieu Hoang
8f272e04a9
output debugging messages to stderr, not stdout
2015-05-05 12:01:21 +04:00
Hieu Hoang
d456d9229e
add multi-bleu-detok. Like multi-bleu scoring but will detokenize/post-process before scoring
2015-05-03 14:07:12 +04:00
Philipp Koehn
a4a7c14593
allow breaking up training data for fast align (to avoid memory blowups for very large corpora)
2015-05-01 17:47:08 -04:00
Philipp Koehn
b369699661
various small changes, mostly related to better compliance with grid engine
2015-05-01 17:44:18 -04:00
Rico Sennrich
e98a2fc980
fix interpolation for LM with parser in pre-processing
2015-04-30 15:46:33 +01:00
Hieu Hoang
4b47e1148c
use ignore-unless /Philipp Koehn
2015-04-22 23:02:57 +04:00
Hieu Hoang
40933b4a78
hack to allow target side of tokenized parallel corpus to be used for LM
2015-04-22 19:01:12 +04:00
Hieu Hoang
ab01d30687
make sure GetOptions doesn consume -T by confusing it with --text
2015-04-21 17:53:46 +04:00
Rico Sennrich
15d3c3f259
be more tolerant about xml input
2015-04-21 14:04:25 +01:00
Rico Sennrich
5a3d5b6bdd
EMS: LM:mock-parse can be actual parser
2015-04-21 10:21:24 +01:00
Hieu Hoang
1b9dc6cfae
more butinah tweaks
2015-04-19 11:50:50 +04:00
Hieu Hoang
637e8a17e8
add pre tokenization cleaning script. In case training has bad, overlying long lines which blows up some taggers/segmenters, eg. mada
2015-04-19 11:21:07 +04:00
Hieu Hoang
6162223690
add use warnings to all perl scripts
2015-04-13 20:42:33 +04:00
Dingyuan Wang
4aba64ed53
Merge pull request #106 from gumblex/master
...
Fix some problems in EMS
2015-04-11 09:26:25 +08:00
Hieu Hoang
02185a85fb
store temp run files in current directory, not /tmp
2015-04-05 17:02:48 +04:00
Hieu Hoang
93ad52d2f9
leave in runPath for debugging
2015-04-05 16:49:12 +04:00
Hieu Hoang
7ffdddef13
script to submit ems job to grid engine as 1 job. Hardcoded for NYUAD at the mo
2015-04-05 16:44:24 +04:00
Dingyuan Wang
aea07b0a19
Fix some problems in EMS:
...
* remove absolute links
* fix coverage bar highlighting
* change Base64 library to support UTF-8
2015-04-03 23:47:25 +08:00
Hieu Hoang
b2f9ba2b64
revert last commit to add MASTER_PATH. Not needed
2015-04-02 19:29:42 +04:00
Hieu Hoang
27b36e0c96
pass in PATH variable from master node. When you're running of a grid but really just qsubbing everything to 1 slave node
2015-04-02 19:15:21 +04:00
Hieu Hoang
2d1da3219d
consistently use 'env perl' command for environments where the 1st perl in PATH isn't the default perl. Which is kinda stupid
2015-04-02 17:38:56 +04:00
Hieu Hoang
e22d275c32
don't ignore lowercasing of factored LM. Must be consistent with pt
2015-04-01 23:25:57 +04:00
Phil Williams
6ce3060dd8
lmplz-wrapper.perl: use Getopt::Long's "pass_through" option
...
This avoids the need to duplicate all of lmplz's options in the wrapper and
it prevents --prune 0 0 1 from being truncated to --prune 0 if the user forgets
to quote the arguments.
2015-03-30 10:18:51 +01:00
Rico Sennrich
3a673fc8dc
EMS: support for syntactic metrics for MERT/MIRA
...
- add "-n-best-trees" to TUNING:decoder-settings
- add "mock-output-parser-references = $output-parser" to GENERAL (and define output-parser)
- TUNING:tuning-settings should include the metric you want to optimize (e.g. "-batch-mira-args='--sctype BLEU,HWCM'")
2015-03-20 17:15:33 +00:00
Phil Williams
fc15e03ebe
Replace truecase-egret.sh with more general tree-converter-wrapper.perl
2015-03-18 09:57:42 +00:00
Phil Williams
0a8e5fb3bf
EMS: fix TRAINING:use-syntax-input-weight-feature option
2015-03-13 17:18:56 +00:00
Hieu Hoang
ce8b0e0876
fix example for reusing tuned moses.ini file
2015-03-13 15:07:23 +00:00
Philipp Koehn
530d0f5a11
some more better defaults for recaser
2015-03-11 17:56:02 +00:00
Philipp Koehn
2ce45229f8
better default configuration for recaser
2015-03-11 17:52:30 +00:00
Philipp Koehn
1632c5f39d
proper handling of specified configuration file
2015-03-11 16:49:20 +00:00
Matthias Huck
01bed83cf9
GHKM extraction: option to strip non-terminal labels from BitPar syntactic parses right during extraction (i.e., remove any suffix starting with a hyphen from the label)
2015-03-10 21:25:32 +00:00
Phil Williams
9e2eb702dc
EMS: add TRAINING:use-syntax-input-weight-feature option
2015-03-10 11:40:49 +00:00
Phil Williams
7eba58b942
EMS: add TRAINING:dont-tune-glue-grammar option
...
Adds -dont-tune-glue-grammar to train-model.perl command during config file
generation step. This is preferable to manually adding -dont-tune-glue-grammar
to TRAINING:training-options because changing its value won't trigger a re-run
of dependent steps that don't really need re-running (like word alignment).
2015-03-10 10:20:19 +00:00