davidstap commited on
Commit
91ac2f8
·
1 Parent(s): 8d3200c

add script

Browse files
Files changed (1) hide show
  1. ted_talks.py +153 -0
ted_talks.py ADDED
@@ -0,0 +1,153 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import datasets
2
+
3
+
4
+ _DESCRIPTION = """\
5
+ Train, validation and test splits for TED talks as in http://phontron.com/data/ted_talks.tar.gz (detokenized)
6
+ """
7
+
8
+ _CITATION = """\
9
+ @inproceedings{Ye2018WordEmbeddings,
10
+ author = {Ye, Qi and Devendra, Sachan and Matthieu, Felix and Sarguna, Padmanabhan and Graham, Neubig},
11
+ title = {When and Why are pre-trained word embeddings useful for Neural Machine Translation},
12
+ booktitle = {HLT-NAACL},
13
+ year = {2018},
14
+ }
15
+ """
16
+
17
+ _DATA_URL = "data/TED.tar"
18
+
19
+ _LANGUAGES = ["ar", "az", "be", "bg", "bn", "bs", "cs", "da", "de", "el", "en", "eo", "es", "et", "eu", "fa", "fi", "fr", "fr-ca", "gl", "he", "hi", "hr", "hu", "hy", "id", "it", "ja", "ka", "kk", "ko", "ku", "lt", "mk", "mn", "mr", "ms", "my", "nb", "nl", "pl", "pt", "pt-br", "ro", "ru", "sk", "sl", "sq", "sr", "sv", "ta", "th", "tr", "uk", "ur", "vi", "zh", "zh-cn", "zh-tw"]
20
+
21
+
22
+ class TedTalksConfig(datasets.BuilderConfig):
23
+ """BuilderConfig for TED talk dataset."""
24
+
25
+ def __init__(self, language_pair=(None, None), **kwargs):
26
+ # sort such that az_ar is same as ar_az
27
+ self.language_pair = sorted(language_pair)
28
+ self.source, self.target = self.language_pair[0], self.language_pair[1]
29
+
30
+ name = f"{self.source}_{self.target}"
31
+ description = f"Parallel sentences in `{self.source}` and `{self.target}`."
32
+ super(TedTalksConfig, self).__init__(name=name, description=description, **kwargs)
33
+
34
+
35
+ class TedTalks(datasets.GeneratorBasedBuilder):
36
+ """TED talk data from http://phontron.com/data/ted_talks.tar.gz."""
37
+
38
+ unique_pairs = sorted(set([
39
+ "_".join(sorted([l1, l2]))
40
+ for l1 in _LANGUAGES
41
+ for l2 in _LANGUAGES
42
+ if l1 != l2
43
+ ]))
44
+
45
+ BUILDER_CONFIGS = [
46
+ TedTalksConfig(
47
+ language_pair=(pair.split("_")[0], pair.split("_")[1]),
48
+ version=datasets.Version("1.0.0", ""),
49
+ )
50
+ for pair in unique_pairs
51
+ ]
52
+
53
+ def _info(self):
54
+ return datasets.DatasetInfo(
55
+ description=_DESCRIPTION,
56
+ features=datasets.Features(
57
+ {"translation": datasets.features.Translation(languages=self.config.language_pair)}
58
+ ),
59
+ homepage="https://github.com/neulab/word-embeddings-for-nmt",
60
+ citation=_CITATION,
61
+ )
62
+
63
+ def _split_generators(self, dl_manager):
64
+ archive = dl_manager.download(_DATA_URL)
65
+
66
+ def _get_overlap(source_file, target_file):
67
+ for path, f in dl_manager.iter_archive(archive):
68
+ if path == source_file:
69
+ source_sentences = f.read().decode("utf-8").split("\n")
70
+ elif path == target_file:
71
+ target_sentences = f.read().decode("utf-8").split("\n")
72
+
73
+ return len([
74
+ (src, tgt)
75
+ for src, tgt
76
+ in zip(source_sentences, target_sentences)
77
+ if src != "" and tgt != ""
78
+ ])
79
+
80
+ split2tedsplit = {"train": "train", "validation": "dev", "test": "test"}
81
+
82
+ overlap = {
83
+ split: _get_overlap(
84
+ f"{split}/ted.{split2tedsplit[split]}.{self.config.source}",
85
+ f"{split}/ted.{split2tedsplit[split]}.{self.config.target}"
86
+ ) for split in ["train", "validation", "test"]
87
+ }
88
+
89
+ generators = []
90
+ if overlap["train"] > 0:
91
+ generators.append(
92
+ datasets.SplitGenerator(
93
+ name=datasets.Split.TRAIN,
94
+ gen_kwargs={
95
+ "source_file": f"train/ted.train.{self.config.source}",
96
+ "target_file": f"train/ted.train.{self.config.target}",
97
+ "files": dl_manager.iter_archive(archive),
98
+ },
99
+ ),
100
+ )
101
+ if overlap["validation"] > 0:
102
+ generators.append(
103
+ datasets.SplitGenerator(
104
+ name=datasets.Split.VALIDATION,
105
+ gen_kwargs={
106
+ "source_file": f"validation/ted.dev.{self.config.source}",
107
+ "target_file": f"validation/ted.dev.{self.config.target}",
108
+ "files": dl_manager.iter_archive(archive),
109
+ },
110
+ ),
111
+ )
112
+ if overlap["test"] > 0:
113
+ generators.append(
114
+ datasets.SplitGenerator(
115
+ name=datasets.Split.TEST,
116
+ gen_kwargs={
117
+ "source_file": f"test/ted.test.{self.config.source}",
118
+ "target_file": f"test/ted.test.{self.config.target}",
119
+ "files": dl_manager.iter_archive(archive),
120
+ },
121
+ ),
122
+ )
123
+
124
+ return generators
125
+
126
+
127
+ def _generate_examples(self, source_file, target_file, files):
128
+ """Returns examples as raw text."""
129
+
130
+ source_sentences, target_sentences = None, None
131
+ for path, f in files:
132
+ if path == source_file:
133
+ source_sentences = f.read().decode("utf-8").split("\n")
134
+ elif path == target_file:
135
+ target_sentences = f.read().decode("utf-8").split("\n")
136
+
137
+ assert len(target_sentences) == len(source_sentences), (
138
+ f"Sizes do not match: {len(source_sentences)} vs {len(target_sentences)}."
139
+ )
140
+
141
+ # ignore empty
142
+ source_target_pairs = [
143
+ (src, tgt)
144
+ for src, tgt
145
+ in zip(source_sentences, target_sentences)
146
+ if src != "" and tgt != ""
147
+ ]
148
+
149
+ if len(source_target_pairs) > 0:
150
+ source_sentences, target_sentences = zip(*source_target_pairs)
151
+
152
+ for idx, (l1, l2) in enumerate(zip(source_sentences, target_sentences)):
153
+ yield idx, {"translation": {self.config.source: l1, self.config.target: l2}}