blob: 62da8dd257145091289c4d71dd59c9d53b89d8a4 (
plain)
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
|
#!/bin/bash -v
MARIAN=../..
# set chosen gpus
GPUS=0
if [ $# -ne 0 ]
then
GPUS=$@
fi
echo Using GPUs: $GPUS
if [ ! -e $MARIAN/build/marian ]
then
echo "marian is not installed in $MARIAN/build, you need to compile the toolkit first"
exit 1
fi
# get our fork of sacrebleu
git clone https://github.com/marian-nmt/sacreBLEU
if [ ! -e "data/corpus.ro" ]
then
cd data
# create dev set
../sacreBLEU/sacrebleu.py -t wmt16/dev -l ro-en --echo src > newsdev2016.ro
../sacreBLEU/sacrebleu.py -t wmt16/dev -l ro-en --echo ref > newsdev2016.en
# create test set
../sacreBLEU/sacrebleu.py -t wmt16 -l ro-en --echo src > newstest2016.ro
../sacreBLEU/sacrebleu.py -t wmt16 -l ro-en --echo ref > newstest2016.en
# get En-Ro training data for WMT16
wget -nc http://www.statmt.org/europarl/v7/ro-en.tgz
wget -nc http://opus.lingfil.uu.se/download.php?f=SETIMES2/en-ro.txt.zip -O SETIMES2.ro-en.txt.zip
wget -nc http://data.statmt.org/rsennrich/wmt16_backtranslations/ro-en/corpus.bt.ro-en.en.gz
wget -nc http://data.statmt.org/rsennrich/wmt16_backtranslations/ro-en/corpus.bt.ro-en.ro.gz
# extract data
tar -xf ro-en.tgz
unzip SETIMES2.ro-en.txt.zip
gzip -d corpus.bt.ro-en.en.gz corpus.bt.ro-en.ro.gz
# create corpus files
cat europarl-v7.ro-en.en SETIMES2.en-ro.en corpus.bt.ro-en.en > corpus.en
cat europarl-v7.ro-en.ro SETIMES2.en-ro.ro corpus.bt.ro-en.ro > corpus.ro
# clean
rm ro-en.tgz SETIMES2.* corpus.bt.* europarl-*
cd ..
fi
# create the model folder
mkdir -p model
# train model
if [ ! -e "model/model.npz.best-bleu-detok.npz" ]
then
$MARIAN/build/marian \
--devices $GPUS \
--type s2s \
--model model/model.npz \
--train-sets data/corpus.ro data/corpus.en \
--vocabs model/vocab.roen.spm model/vocab.roen.spm \
--sentencepiece-options '--normalization_rule_tsv=data/norm_romanian.tsv' \
--dim-vocabs 32000 32000 \
--mini-batch-fit -w 5000 \
--layer-normalization --tied-embeddings-all \
--dropout-rnn 0.2 --dropout-src 0.1 --dropout-trg 0.1 \
--early-stopping 5 --max-length 100 \
--valid-freq 10000 --save-freq 10000 --disp-freq 1000 \
--cost-type ce-mean-words --valid-metrics ce-mean-words bleu-detok \
--valid-sets data/newsdev2016.ro data/newsdev2016.en \
--log model/train.log --valid-log model/valid.log --tempdir model \
--overwrite --keep-best \
--seed 1111 --exponential-smoothing \
--normalize=1 --beam-size=6 --quiet-translation
fi
# translate dev set
cat data/newsdev2016.ro \
| $MARIAN/build/marian-decoder -c model/model.npz.best-bleu-detok.npz.decoder.yml -d $GPUS -b 6 -n1 \
--mini-batch 64 --maxi-batch 10 --maxi-batch-sort src > data/newsdev2016.ro.output
# translate test set
cat data/newstest2016.ro \
| $MARIAN/build/marian-decoder -c model/model.npz.best-bleu-detok.npz.decoder.yml -d $GPUS -b 6 -n1 \
--mini-batch 64 --maxi-batch 10 --maxi-batch-sort src > data/newstest2016.ro.output
# calculate bleu scores on dev and test set
sacreBLEU/sacrebleu.py -t wmt16/dev -l ro-en < data/newsdev2016.ro.output
sacreBLEU/sacrebleu.py -t wmt16 -l ro-en < data/newstest2016.ro.output
|
