Datasets:
GEM
/

License:
j-chim commited on
Commit
550c363
1 Parent(s): 78a734e

update per-language splits

Browse files
Files changed (1) hide show
  1. wiki_lingua.py +46 -41
wiki_lingua.py CHANGED
@@ -17,7 +17,6 @@
17
  import os
18
  import glob
19
  import pickle
20
- import re
21
  import datasets
22
 
23
 
@@ -170,7 +169,7 @@ class WikiLingua(datasets.GeneratorBasedBuilder):
170
  "target_language": datasets.Value("string"),
171
  "source": datasets.Value("string"),
172
  "target": datasets.Value("string"),
173
- "references": [datasets.Value("string")]
174
  }
175
  ),
176
  supervised_keys=None,
@@ -181,59 +180,65 @@ class WikiLingua(datasets.GeneratorBasedBuilder):
181
 
182
  def _split_generators(self, dl_manager):
183
  """Returns SplitGenerators."""
 
184
  dl_dir = dl_manager.download_and_extract(_URL)
185
  data_dir = os.path.join(dl_dir, "GEM_V2")
186
 
187
  return [
188
  datasets.SplitGenerator(
189
  name=datasets.Split.TRAIN,
190
- gen_kwargs={"filepath": data_dir, "split": "train"},
 
 
 
 
191
  ),
192
  datasets.SplitGenerator(
193
  name=datasets.Split.VALIDATION,
194
- gen_kwargs={"filepath": data_dir, "split": "validation"},
 
 
 
 
195
  ),
196
- ] + [
197
  datasets.SplitGenerator(
198
- name=re.search("wikilingua_(.+test)", fn).group(1).replace(".", "_"),
199
- gen_kwargs={"filepath": fn, "split": "test"},
200
- )
201
- for fn in glob.glob(os.path.join(data_dir, "wikilingua_*.test.pk"))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
202
  ]
203
 
204
- def _generate_examples(self, filepath, split):
205
  """Yields examples."""
206
-
207
- if split == "test":
208
- with open(filepath, "rb") as f:
209
- data = pickle.load(f)
210
- for d in data:
211
- idx = d["id"].replace(".", "-")
212
- yield idx, {
213
- "gem_id": idx,
214
- "gem_parent_id": idx,
215
- "source_language": d["source"],
216
- "target_language": d["target"],
217
- "source": d["document"].strip(),
218
- "target": d["summary"].strip(),
219
- "references": [d["summary"].strip()],
220
- }
221
- else:
222
- # filter data as needed for train & validation sets
223
- if split == "validation":
224
- filepaths = glob.glob(os.path.join(filepath, "wikilingua_*.val.pk"))
225
  else:
226
- filepaths = glob.glob(os.path.join(filepath, "wikilingua_*.train.pk"))
227
- for filepath in filepaths:
228
- # skip files if they are irrelevant to task mode
229
- if (
230
- self.config.mode == WikilinguaModes.MULTILINGUAL
231
- and "crosslingual" in filepath
232
- ) or (
233
- self.config.mode == WikilinguaModes.CROSSLINGUAL
234
- and "multilingual" in filepath
235
- ):
236
- yield from []
237
 
238
  with open(filepath, "rb") as f:
239
  data = pickle.load(f)
@@ -265,5 +270,5 @@ class WikiLingua(datasets.GeneratorBasedBuilder):
265
  "target_language": tgt_lang,
266
  "source": src,
267
  "target": tgt,
268
- "references": [tgt]
269
  }
 
17
  import os
18
  import glob
19
  import pickle
 
20
  import datasets
21
 
22
 
 
169
  "target_language": datasets.Value("string"),
170
  "source": datasets.Value("string"),
171
  "target": datasets.Value("string"),
172
+ "references": [datasets.Value("string")],
173
  }
174
  ),
175
  supervised_keys=None,
 
180
 
181
  def _split_generators(self, dl_manager):
182
  """Returns SplitGenerators."""
183
+
184
  dl_dir = dl_manager.download_and_extract(_URL)
185
  data_dir = os.path.join(dl_dir, "GEM_V2")
186
 
187
  return [
188
  datasets.SplitGenerator(
189
  name=datasets.Split.TRAIN,
190
+ gen_kwargs={
191
+ "filepaths": glob.glob(
192
+ os.path.join(data_dir, f"wikilingua_*.train.pk")
193
+ )
194
+ },
195
  ),
196
  datasets.SplitGenerator(
197
  name=datasets.Split.VALIDATION,
198
+ gen_kwargs={
199
+ "filepaths": glob.glob(
200
+ os.path.join(data_dir, f"wikilingua_*lingual.val.pk")
201
+ )
202
+ },
203
  ),
 
204
  datasets.SplitGenerator(
205
+ name=datasets.Split.TEST,
206
+ gen_kwargs={
207
+ "filepaths": glob.glob(
208
+ os.path.join(data_dir, f"wikilingua_*lingual.test.pk")
209
+ )
210
+ },
211
+ ),
212
+ datasets.SplitGenerator(
213
+ name=f"sampled_{datasets.Split.VALIDATION}",
214
+ gen_kwargs={
215
+ "filepaths": glob.glob(
216
+ os.path.join(data_dir, f"wikilingua_*_sampled.val.pk")
217
+ )
218
+ },
219
+ ),
220
+ datasets.SplitGenerator(
221
+ name=f"sampled_{datasets.Split.TEST}",
222
+ gen_kwargs={
223
+ "filepaths": glob.glob(
224
+ os.path.join(data_dir, f"wikilingua_*_sampled.test.pk")
225
+ )
226
+ },
227
+ ),
228
  ]
229
 
230
+ def _generate_examples(self, filepaths):
231
  """Yields examples."""
232
+ for filepath in filepaths:
233
+ if (
234
+ self.config.name == WikilinguaModes.MULTILINGUAL
235
+ and WikilinguaModes.CROSSLINGUAL in filepath
236
+ ) or (
237
+ self.config.name == WikilinguaModes.CROSSLINGUAL
238
+ and WikilinguaModes.MULTILINGUAL in filepath
239
+ ):
240
+ yield from []
 
 
 
 
 
 
 
 
 
 
241
  else:
 
 
 
 
 
 
 
 
 
 
 
242
 
243
  with open(filepath, "rb") as f:
244
  data = pickle.load(f)
 
270
  "target_language": tgt_lang,
271
  "source": src,
272
  "target": tgt,
273
+ "references": [tgt],
274
  }