Datasets:

Multilinguality:
translation
Size Categories:
1M<n<10M
Language Creators:
expert-generated
Annotations Creators:
crowdsourced
Source Datasets:
original
Tags:
License:
albertvillanova HF staff commited on
Commit
49aad53
1 Parent(s): b33c264

Support streaming iwslt2017 dataset (#4992)

Browse files

Commit from https://github.com/huggingface/datasets/commit/5b7cc0b1ceae98b9791e9df102ee279fabedd469

Files changed (1) hide show
  1. iwslt2017.py +12 -21
iwslt2017.py CHANGED
@@ -20,6 +20,12 @@ import os
20
  import datasets
21
 
22
 
 
 
 
 
 
 
23
  _CITATION = """\
24
  @inproceedings{cettolo-etal-2017-overview,
25
  title = "Overview of the {IWSLT} 2017 Evaluation Campaign",
@@ -41,11 +47,9 @@ _CITATION = """\
41
  }
42
  """
43
 
44
- _DESCRIPTION = """\
45
- The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.
46
- """
47
-
48
- MULTI_URL = "https://huggingface.co/datasets/iwslt2017/resolve/ebd7c60d9800c2a1be010a227e5f0a2363730f7a/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.tgz"
49
 
50
 
51
  class IWSLT2017Config(datasets.BuilderConfig):
@@ -100,18 +104,11 @@ class IWSLT217(datasets.GeneratorBasedBuilder):
100
 
101
  def _info(self):
102
  return datasets.DatasetInfo(
103
- # This is the description that will appear on the datasets page.
104
  description=_DESCRIPTION,
105
- # datasets.features.FeatureConnectors
106
  features=datasets.Features(
107
  {"translation": datasets.features.Translation(languages=self.config.pair.split("-"))}
108
  ),
109
- # If there's a common (input, target) tuple from the features,
110
- # specify them here. They'll be used if as_supervised=True in
111
- # builder.as_dataset.
112
- supervised_keys=None,
113
- # Homepage of the dataset for documentation
114
- homepage="https://sites.google.com/site/iwsltevaluation2017/TED-tasks",
115
  citation=_CITATION,
116
  )
117
 
@@ -123,14 +120,13 @@ class IWSLT217(datasets.GeneratorBasedBuilder):
123
  data_dir = os.path.join(dl_dir, "DeEnItNlRo-DeEnItNlRo")
124
  years = [2010]
125
  else:
126
- bi_url = f"https://huggingface.co/datasets/iwslt2017/resolve/ebd7c60d9800c2a1be010a227e5f0a2363730f7a/data/2017-01-trnted/texts/{source}/{target}/{source}-{target}.tgz"
127
  dl_dir = dl_manager.download_and_extract(bi_url)
128
  data_dir = os.path.join(dl_dir, f"{source}-{target}")
129
  years = [2010, 2011, 2012, 2013, 2014, 2015]
130
  return [
131
  datasets.SplitGenerator(
132
  name=datasets.Split.TRAIN,
133
- # These kwargs will be passed to _generate_examples
134
  gen_kwargs={
135
  "source_files": [
136
  os.path.join(
@@ -144,12 +140,10 @@ class IWSLT217(datasets.GeneratorBasedBuilder):
144
  f"train.tags.{self.config.pair}.{target}",
145
  )
146
  ],
147
- "split": "train",
148
  },
149
  ),
150
  datasets.SplitGenerator(
151
  name=datasets.Split.TEST,
152
- # These kwargs will be passed to _generate_examples
153
  gen_kwargs={
154
  "source_files": [
155
  os.path.join(
@@ -165,12 +159,10 @@ class IWSLT217(datasets.GeneratorBasedBuilder):
165
  )
166
  for year in years
167
  ],
168
- "split": "test",
169
  },
170
  ),
171
  datasets.SplitGenerator(
172
  name=datasets.Split.VALIDATION,
173
- # These kwargs will be passed to _generate_examples
174
  gen_kwargs={
175
  "source_files": [
176
  os.path.join(
@@ -184,12 +176,11 @@ class IWSLT217(datasets.GeneratorBasedBuilder):
184
  f"IWSLT17.TED.dev2010.{self.config.pair}.{target}.xml",
185
  )
186
  ],
187
- "split": "dev",
188
  },
189
  ),
190
  ]
191
 
192
- def _generate_examples(self, source_files, target_files, split):
193
  """Yields examples."""
194
  id_ = 0
195
  source, target = self.config.pair.split("-")
20
  import datasets
21
 
22
 
23
+ _HOMEPAGE = "https://sites.google.com/site/iwsltevaluation2017/TED-tasks"
24
+
25
+ _DESCRIPTION = """\
26
+ The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.
27
+ """
28
+
29
  _CITATION = """\
30
  @inproceedings{cettolo-etal-2017-overview,
31
  title = "Overview of the {IWSLT} 2017 Evaluation Campaign",
47
  }
48
  """
49
 
50
+ REPO_URL = "https://huggingface.co/datasets/iwslt2017/resolve/main/"
51
+ MULTI_URL = REPO_URL + "data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip"
52
+ BI_URL = REPO_URL + "data/2017-01-trnted/texts/{source}/{target}/{source}-{target}.zip"
 
 
53
 
54
 
55
  class IWSLT2017Config(datasets.BuilderConfig):
104
 
105
  def _info(self):
106
  return datasets.DatasetInfo(
 
107
  description=_DESCRIPTION,
 
108
  features=datasets.Features(
109
  {"translation": datasets.features.Translation(languages=self.config.pair.split("-"))}
110
  ),
111
+ homepage=_HOMEPAGE,
 
 
 
 
 
112
  citation=_CITATION,
113
  )
114
 
120
  data_dir = os.path.join(dl_dir, "DeEnItNlRo-DeEnItNlRo")
121
  years = [2010]
122
  else:
123
+ bi_url = BI_URL.format(source=source, target=target)
124
  dl_dir = dl_manager.download_and_extract(bi_url)
125
  data_dir = os.path.join(dl_dir, f"{source}-{target}")
126
  years = [2010, 2011, 2012, 2013, 2014, 2015]
127
  return [
128
  datasets.SplitGenerator(
129
  name=datasets.Split.TRAIN,
 
130
  gen_kwargs={
131
  "source_files": [
132
  os.path.join(
140
  f"train.tags.{self.config.pair}.{target}",
141
  )
142
  ],
 
143
  },
144
  ),
145
  datasets.SplitGenerator(
146
  name=datasets.Split.TEST,
 
147
  gen_kwargs={
148
  "source_files": [
149
  os.path.join(
159
  )
160
  for year in years
161
  ],
 
162
  },
163
  ),
164
  datasets.SplitGenerator(
165
  name=datasets.Split.VALIDATION,
 
166
  gen_kwargs={
167
  "source_files": [
168
  os.path.join(
176
  f"IWSLT17.TED.dev2010.{self.config.pair}.{target}.xml",
177
  )
178
  ],
 
179
  },
180
  ),
181
  ]
182
 
183
+ def _generate_examples(self, source_files, target_files):
184
  """Yields examples."""
185
  id_ = 0
186
  source, target = self.config.pair.split("-")