-
Notifications
You must be signed in to change notification settings - Fork 2
/
get_data_enfr.sh
executable file
·302 lines (253 loc) · 10.6 KB
/
get_data_enfr.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
# Copyright (c) 2018-present, Facebook, Inc.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
#
set -e
#
# Data preprocessing configuration
#
N_MONO=10000000 # number of monolingual sentences for each language
CODES=60000 # number of BPE codes
N_THREADS=48 # number of threads in data preprocessing
N_EPOCHS=10 # number of fastText epochs
#
# Initialize tools and data paths
#
# main paths
UMT_PATH=$PWD
TOOLS_PATH=$PWD/tools
DATA_PATH=$PWD/data
MONO_PATH=$DATA_PATH/mono
PARA_PATH=$DATA_PATH/para
# create paths
mkdir -p $TOOLS_PATH
mkdir -p $DATA_PATH
mkdir -p $MONO_PATH
mkdir -p $PARA_PATH
# moses
MOSES=$TOOLS_PATH/mosesdecoder
TOKENIZER=$MOSES/scripts/tokenizer/tokenizer.perl
NORM_PUNC=$MOSES/scripts/tokenizer/normalize-punctuation.perl
INPUT_FROM_SGM=$MOSES/scripts/ems/support/input-from-sgm.perl
REM_NON_PRINT_CHAR=$MOSES/scripts/tokenizer/remove-non-printing-char.perl
# fastBPE
FASTBPE_DIR=$TOOLS_PATH/fastBPE
FASTBPE=$FASTBPE_DIR/fast
# fastText
FASTTEXT_DIR=$TOOLS_PATH/fastText
FASTTEXT=$FASTTEXT_DIR/fasttext
# files full paths
SRC_RAW=$MONO_PATH/all.en
TGT_RAW=$MONO_PATH/all.fr
SRC_TOK=$MONO_PATH/all.en.tok
TGT_TOK=$MONO_PATH/all.fr.tok
BPE_CODES=$MONO_PATH/bpe_codes
CONCAT_BPE=$MONO_PATH/all.en-fr.$CODES
SRC_VOCAB=$MONO_PATH/vocab.en.$CODES
TGT_VOCAB=$MONO_PATH/vocab.fr.$CODES
FULL_VOCAB=$MONO_PATH/vocab.en-fr.$CODES
SRC_VALID=$PARA_PATH/dev/newstest2013-ref.en
TGT_VALID=$PARA_PATH/dev/newstest2013-ref.fr
SRC_TEST=$PARA_PATH/dev/newstest2014-fren-src.en
TGT_TEST=$PARA_PATH/dev/newstest2014-fren-src.fr
#
# Download and install tools
#
# Download Moses
cd $TOOLS_PATH
if [ ! -d "$MOSES" ]; then
echo "Cloning Moses from GitHub repository..."
git clone https://github.com/moses-smt/mosesdecoder.git
fi
echo "Moses found in: $MOSES"
# Download fastBPE
cd $TOOLS_PATH
if [ ! -d "$FASTBPE_DIR" ]; then
echo "Cloning fastBPE from GitHub repository..."
git clone https://github.com/glample/fastBPE
fi
echo "fastBPE found in: $FASTBPE_DIR"
# Compile fastBPE
cd $TOOLS_PATH
if [ ! -f "$FASTBPE" ]; then
echo "Compiling fastBPE..."
cd $FASTBPE_DIR
g++ -std=c++11 -pthread -O3 fast.cc -o fast
fi
echo "fastBPE compiled in: $FASTBPE"
# Download fastText
cd $TOOLS_PATH
if [ ! -d "$FASTTEXT_DIR" ]; then
echo "Cloning fastText from GitHub repository..."
git clone https://github.com/facebookresearch/fastText.git
fi
echo "fastText found in: $FASTTEXT_DIR"
# Compile fastText
cd $TOOLS_PATH
if [ ! -f "$FASTTEXT" ]; then
echo "Compiling fastText..."
cd $FASTTEXT_DIR
make
fi
echo "fastText compiled in: $FASTTEXT"
#
# Download monolingual data
#
cd $MONO_PATH
echo "Downloading English files..."
wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2007.en.shuffled.gz
wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2008.en.shuffled.gz
wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2009.en.shuffled.gz
wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2010.en.shuffled.gz
# wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2011.en.shuffled.gz
# wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2012.en.shuffled.gz
# wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2013.en.shuffled.gz
# wget -c http://www.statmt.org/wmt15/training-monolingual-news-crawl-v2/news.2014.en.shuffled.v2.gz
# wget -c http://data.statmt.org/wmt16/translation-task/news.2015.en.shuffled.gz
# wget -c http://data.statmt.org/wmt17/translation-task/news.2016.en.shuffled.gz
# wget -c http://data.statmt.org/wmt18/translation-task/news.2017.en.shuffled.deduped.gz
echo "Downloading French files..."
wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2007.fr.shuffled.gz
wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2008.fr.shuffled.gz
wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2009.fr.shuffled.gz
wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2010.fr.shuffled.gz
# wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2011.fr.shuffled.gz
# wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2012.fr.shuffled.gz
# wget -c http://www.statmt.org/wmt14/training-monolingual-news-crawl/news.2013.fr.shuffled.gz
# wget -c http://www.statmt.org/wmt15/training-monolingual-news-crawl-v2/news.2014.fr.shuffled.v2.gz
# wget -c http://data.statmt.org/wmt17/translation-task/news.2015.fr.shuffled.gz
# wget -c http://data.statmt.org/wmt17/translation-task/news.2016.fr.shuffled.gz
# wget -c http://data.statmt.org/wmt17/translation-task/news.2017.fr.shuffled.gz
# decompress monolingual data
for FILENAME in news*gz; do
OUTPUT="${FILENAME::-3}"
if [ ! -f "$OUTPUT" ]; then
echo "Decompressing $FILENAME..."
# gunzip -c $FILENAME > $OUTPUT # for gunzip version < 1.6
gunzip -k $FILENAME
else
echo "$OUTPUT already decompressed."
fi
done
# concatenate monolingual data files
if ! [[ -f "$SRC_RAW" && -f "$TGT_RAW" ]]; then
echo "Concatenating monolingual data..."
cat $(ls news*en* | grep -v gz) | head -n $N_MONO > $SRC_RAW
cat $(ls news*fr* | grep -v gz) | head -n $N_MONO > $TGT_RAW
fi
echo "EN monolingual data concatenated in: $SRC_RAW"
echo "FR monolingual data concatenated in: $TGT_RAW"
# check number of lines
if ! [[ "$(wc -l < $SRC_RAW)" -eq "$N_MONO" ]]; then echo "ERROR: Number of lines doesn't match! Be sure you have $N_MONO sentences in your EN monolingual data."; exit; fi
if ! [[ "$(wc -l < $TGT_RAW)" -eq "$N_MONO" ]]; then echo "ERROR: Number of lines doesn't match! Be sure you have $N_MONO sentences in your FR monolingual data."; exit; fi
# tokenize data
if ! [[ -f "$SRC_TOK" && -f "$TGT_TOK" ]]; then
echo "Tokenize monolingual data..."
cat $SRC_RAW | $NORM_PUNC -l en | $TOKENIZER -l en -no-escape -threads $N_THREADS > $SRC_TOK
cat $TGT_RAW | $NORM_PUNC -l fr | $TOKENIZER -l fr -no-escape -threads $N_THREADS > $TGT_TOK
fi
echo "EN monolingual data tokenized in: $SRC_TOK"
echo "FR monolingual data tokenized in: $TGT_TOK"
# learn BPE codes
if [ ! -f "$BPE_CODES" ]; then
echo "Learning BPE codes..."
$FASTBPE learnbpe $CODES $SRC_TOK $TGT_TOK > $BPE_CODES
fi
echo "BPE learned in $BPE_CODES"
# apply BPE codes
if ! [[ -f "$SRC_TOK.$CODES" && -f "$TGT_TOK.$CODES" ]]; then
echo "Applying BPE codes..."
$FASTBPE applybpe $SRC_TOK.$CODES $SRC_TOK $BPE_CODES
$FASTBPE applybpe $TGT_TOK.$CODES $TGT_TOK $BPE_CODES
fi
echo "BPE codes applied to EN in: $SRC_TOK.$CODES"
echo "BPE codes applied to FR in: $TGT_TOK.$CODES"
# extract vocabulary
if ! [[ -f "$SRC_VOCAB" && -f "$TGT_VOCAB" && -f "$FULL_VOCAB" ]]; then
echo "Extracting vocabulary..."
$FASTBPE getvocab $SRC_TOK.$CODES > $SRC_VOCAB
$FASTBPE getvocab $TGT_TOK.$CODES > $TGT_VOCAB
$FASTBPE getvocab $SRC_TOK.$CODES $TGT_TOK.$CODES > $FULL_VOCAB
fi
echo "EN vocab in: $SRC_VOCAB"
echo "FR vocab in: $TGT_VOCAB"
echo "Full vocab in: $FULL_VOCAB"
# binarize data
#if ! [[ -f "$SRC_TOK.$CODES.pth" && -f "$TGT_TOK.$CODES.pth" ]]; then
# echo "Binarizing data..."
# $UMT_PATH/preprocess.py $FULL_VOCAB $SRC_TOK.$CODES
# $UMT_PATH/preprocess.py $FULL_VOCAB $TGT_TOK.$CODES
#fi
#echo "EN binarized data in: $SRC_TOK.$CODES.pth"
#echo "FR binarized data in: $TGT_TOK.$CODES.pth"
#
# Download parallel data (for evaluation only)
#
cd $PARA_PATH
echo "Downloading parallel data..."
wget -c http://data.statmt.org/wmt17/translation-task/dev.tgz
echo "Extracting parallel data..."
tar -xzf dev.tgz
# check valid and test files are here
if ! [[ -f "$SRC_VALID.sgm" ]]; then echo "$SRC_VALID.sgm is not found!"; exit; fi
if ! [[ -f "$TGT_VALID.sgm" ]]; then echo "$TGT_VALID.sgm is not found!"; exit; fi
if ! [[ -f "$SRC_TEST.sgm" ]]; then echo "$SRC_TEST.sgm is not found!"; exit; fi
if ! [[ -f "$TGT_TEST.sgm" ]]; then echo "$TGT_TEST.sgm is not found!"; exit; fi
echo "Tokenizing valid and test data..."
$INPUT_FROM_SGM < $SRC_VALID.sgm | $NORM_PUNC -l en | $REM_NON_PRINT_CHAR | $TOKENIZER -l en -no-escape -threads $N_THREADS > $SRC_VALID
$INPUT_FROM_SGM < $TGT_VALID.sgm | $NORM_PUNC -l fr | $REM_NON_PRINT_CHAR | $TOKENIZER -l fr -no-escape -threads $N_THREADS > $TGT_VALID
$INPUT_FROM_SGM < $SRC_TEST.sgm | $NORM_PUNC -l en | $REM_NON_PRINT_CHAR | $TOKENIZER -l en -no-escape -threads $N_THREADS > $SRC_TEST
$INPUT_FROM_SGM < $TGT_TEST.sgm | $NORM_PUNC -l fr | $REM_NON_PRINT_CHAR | $TOKENIZER -l fr -no-escape -threads $N_THREADS > $TGT_TEST
echo "Applying BPE to valid and test files..."
$FASTBPE applybpe $SRC_VALID.$CODES $SRC_VALID $BPE_CODES $SRC_VOCAB
$FASTBPE applybpe $TGT_VALID.$CODES $TGT_VALID $BPE_CODES $TGT_VOCAB
$FASTBPE applybpe $SRC_TEST.$CODES $SRC_TEST $BPE_CODES $SRC_VOCAB
$FASTBPE applybpe $TGT_TEST.$CODES $TGT_TEST $BPE_CODES $TGT_VOCAB
#echo "Binarizing data..."
#rm -f $SRC_VALID.$CODES.pth $TGT_VALID.$CODES.pth $SRC_TEST.$CODES.pth $TGT_TEST.$CODES.pth
#$UMT_PATH/preprocess.py $FULL_VOCAB $SRC_VALID.$CODES
#$UMT_PATH/preprocess.py $FULL_VOCAB $TGT_VALID.$CODES
#$UMT_PATH/preprocess.py $FULL_VOCAB $SRC_TEST.$CODES
#$UMT_PATH/preprocess.py $FULL_VOCAB $TGT_TEST.$CODES
#
# Summary
#
#echo ""
#echo "===== Data summary"
#echo "Monolingual training data:"
#echo " EN: $SRC_TOK.$CODES.pth"
#echo " FR: $TGT_TOK.$CODES.pth"
#echo "Parallel validation data:"
#echo " EN: $SRC_VALID.$CODES.pth"
#echo " FR: $TGT_VALID.$CODES.pth"
#echo "Parallel test data:"
#echo " EN: $SRC_TEST.$CODES.pth"
#echo " FR: $TGT_TEST.$CODES.pth"
#echo ""
#
# Train fastText on concatenated embeddings
#
if ! [[ -f "$CONCAT_BPE" ]]; then
echo "Concatenating source and target monolingual data..."
cat $SRC_TOK.$CODES $TGT_TOK.$CODES | shuf > $CONCAT_BPE
fi
echo "Concatenated data in: $CONCAT_BPE"
if ! [[ -f "$CONCAT_BPE.vec" ]]; then
echo "Training fastText on $CONCAT_BPE..."
$FASTTEXT skipgram -epoch $N_EPOCHS -minCount 0 -dim 512 -thread $N_THREADS -ws 5 -neg 10 -input $CONCAT_BPE -output $CONCAT_BPE
fi
echo "Cross-lingual embeddings in: $CONCAT_BPE.vec"
#
# Prepare vocab file for TensorFlow
#
if ! [[ -f "$SRC_VOCAB.tf" && -f "$TGT_VOCAB.tf" && -f "$FULL_VOCAB.tf" ]]; then
echo "Build vocab files for TensorFLow by removing number and adding <s>, </s>, <unk>, <mask>."
echo -e "<s> 0\n</s> 1\n<unk> 2\n<mask> 3" | cat - $SRC_VOCAB | cut -d' ' -f1 > $SRC_VOCAB.tf
echo -e "<s> 0\n</s> 1\n<unk> 2\n<mask> 3" | cat - $TGT_VOCAB | cut -d' ' -f1 > $TGT_VOCAB.tf
echo -e "<s> 0\n</s> 1\n<unk> 2\n<mask> 3" | cat - $FULL_VOCAB | cut -d' ' -f1 > $FULL_VOCAB.tf
fi
echo "EN TF vocab in: $SRC_VOCAB.tf"
echo "FR TF vocab in: $TGT_VOCAB.tf"
echo "Full TF vocab in: $FULL_VOCAB.tf"