albertvillanova HF staff commited on
Commit
791c5c3
1 Parent(s): 53bd002

Support streaming xtreme dataset for PAWS-X config (#4132)

Browse files

* Support streaming xtreme dataset for PAWS-X config

* Align tasks in dataset card

Commit from https://github.com/huggingface/datasets/commit/8caed0c1e7b9658f08c10c8b90eb203b2cedc8e4

Files changed (2) hide show
  1. README.md +6 -5
  2. xtreme.py +47 -40
README.md CHANGED
@@ -434,6 +434,7 @@ licenses:
434
  multilinguality:
435
  - multilingual
436
  - translation
 
437
  size_categories:
438
  - n<1K
439
  - 1K<n<10K
@@ -449,21 +450,21 @@ source_datasets:
449
  - extended|tatoeba
450
  - extended|squad
451
  task_categories:
 
452
  - question-answering
453
- - structure-prediction
454
  - text-classification
455
  - text-retrieval
 
456
  task_ids:
457
- - open-domain-qa
458
  - multiple-choice-qa
459
  - extractive-qa
460
- - named-entity-recognition
461
- - part-of-speech-tagging
462
  - natural-language-inference
463
  - text-classification-other-paraphrase-identification
464
  - text-retrieval-other-parallel-sentence-retrieval
 
 
465
  paperswithcode_id: xtreme
466
- pretty_name: XTREME
467
  ---
468
 
469
  # Dataset Card for "xtreme"
 
434
  multilinguality:
435
  - multilingual
436
  - translation
437
+ pretty_name: XTREME
438
  size_categories:
439
  - n<1K
440
  - 1K<n<10K
 
450
  - extended|tatoeba
451
  - extended|squad
452
  task_categories:
453
+ - multiple-choice
454
  - question-answering
 
455
  - text-classification
456
  - text-retrieval
457
+ - token-classification
458
  task_ids:
 
459
  - multiple-choice-qa
460
  - extractive-qa
461
+ - open-domain-qa
 
462
  - natural-language-inference
463
  - text-classification-other-paraphrase-identification
464
  - text-retrieval-other-parallel-sentence-retrieval
465
+ - named-entity-recognition
466
+ - part-of-speech-tagging
467
  paperswithcode_id: xtreme
 
468
  ---
469
 
470
  # Dataset Card for "xtreme"
xtreme.py CHANGED
@@ -461,7 +461,6 @@ class Xtreme(datasets.GeneratorBasedBuilder):
461
  ]
462
 
463
  def _info(self):
464
- # TODO(xtreme): Specifies the datasets.DatasetInfo object
465
  features = {text_feature: datasets.Value("string") for text_feature in self.config.text_features.keys()}
466
  if "answers" in features.keys():
467
  features["answers"] = datasets.features.Sequence(
@@ -471,7 +470,7 @@ class Xtreme(datasets.GeneratorBasedBuilder):
471
  }
472
  )
473
  if self.config.name.startswith("PAWS-X"):
474
- features["label"] = datasets.Value("string")
475
  if self.config.name == "XNLI":
476
  features["gold_label"] = datasets.Value("string")
477
 
@@ -517,10 +516,6 @@ class Xtreme(datasets.GeneratorBasedBuilder):
517
 
518
  def _split_generators(self, dl_manager):
519
  """Returns SplitGenerators."""
520
- # TODO(xtreme): Downloads the data and defines the splits
521
- # dl_manager is a datasets.download.DownloadManager that can be used to
522
- # download and extract URLs
523
-
524
  if self.config.name == "tydiqa":
525
  train_url = "v1.1/tydiqa-goldp-v1.1-train.json"
526
  dev_url = "v1.1/tydiqa-goldp-v1.1-dev.json"
@@ -593,30 +588,7 @@ class Xtreme(datasets.GeneratorBasedBuilder):
593
  ),
594
  ]
595
  if self.config.name.startswith("PAWS-X"):
596
- lang = self.config.name.split(".")[1]
597
- paws_x_dir = dl_manager.download_and_extract(self.config.data_url)
598
- data_dir = os.path.join(paws_x_dir, "x-final", lang)
599
- return [
600
- datasets.SplitGenerator(
601
- name=datasets.Split.VALIDATION,
602
- # These kwargs will be passed to _generate_examples
603
- gen_kwargs={"filepath": os.path.join(data_dir, "dev_2k.tsv")},
604
- ),
605
- datasets.SplitGenerator(
606
- name=datasets.Split.TEST,
607
- # These kwargs will be passed to _generate_examples
608
- gen_kwargs={"filepath": os.path.join(data_dir, "test_2k.tsv")},
609
- ),
610
- datasets.SplitGenerator(
611
- name=datasets.Split.TRAIN,
612
- # These kwargs will be passed to _generate_examples
613
- gen_kwargs={
614
- "filepath": os.path.join(data_dir, "translated_train.tsv")
615
- if lang != "en"
616
- else os.path.join(data_dir, "train.tsv")
617
- },
618
- ),
619
- ]
620
  elif self.config.name.startswith("tatoeba"):
621
  lang = self.config.name.split(".")[1]
622
 
@@ -733,16 +705,7 @@ class Xtreme(datasets.GeneratorBasedBuilder):
733
  "gold_label": row["gold_label"],
734
  }
735
  if self.config.name.startswith("PAWS-X"):
736
- with open(filepath, encoding="utf-8") as f:
737
- data = csv.reader(f, delimiter="\t")
738
- next(data) # skip header
739
- for id_, row in enumerate(data):
740
- if len(row) == 4:
741
- yield id_, {
742
- "sentence1": row[1],
743
- "sentence2": row[2],
744
- "label": row[3],
745
- }
746
  if self.config.name.startswith("XQuAD"):
747
  with open(filepath, encoding="utf-8") as f:
748
  xquad = json.load(f)
@@ -857,6 +820,50 @@ class Xtreme(datasets.GeneratorBasedBuilder):
857
  }
858
 
859
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
860
  class UdposParser:
861
 
862
  features = datasets.Features(
 
461
  ]
462
 
463
  def _info(self):
 
464
  features = {text_feature: datasets.Value("string") for text_feature in self.config.text_features.keys()}
465
  if "answers" in features.keys():
466
  features["answers"] = datasets.features.Sequence(
 
470
  }
471
  )
472
  if self.config.name.startswith("PAWS-X"):
473
+ features = PawsxParser.features
474
  if self.config.name == "XNLI":
475
  features["gold_label"] = datasets.Value("string")
476
 
 
516
 
517
  def _split_generators(self, dl_manager):
518
  """Returns SplitGenerators."""
 
 
 
 
519
  if self.config.name == "tydiqa":
520
  train_url = "v1.1/tydiqa-goldp-v1.1-train.json"
521
  dev_url = "v1.1/tydiqa-goldp-v1.1-dev.json"
 
588
  ),
589
  ]
590
  if self.config.name.startswith("PAWS-X"):
591
+ return PawsxParser.split_generators(dl_manager=dl_manager, config=self.config)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
592
  elif self.config.name.startswith("tatoeba"):
593
  lang = self.config.name.split(".")[1]
594
 
 
705
  "gold_label": row["gold_label"],
706
  }
707
  if self.config.name.startswith("PAWS-X"):
708
+ yield from PawsxParser.generate_examples(config=self.config, filepath=filepath, **kwargs)
 
 
 
 
 
 
 
 
 
709
  if self.config.name.startswith("XQuAD"):
710
  with open(filepath, encoding="utf-8") as f:
711
  xquad = json.load(f)
 
820
  }
821
 
822
 
823
+ class PawsxParser:
824
+
825
+ features = datasets.Features(
826
+ {
827
+ "sentence1": datasets.Value("string"),
828
+ "sentence2": datasets.Value("string"),
829
+ "label": datasets.Value("string"),
830
+ }
831
+ )
832
+
833
+ @staticmethod
834
+ def split_generators(dl_manager=None, config=None):
835
+ lang = config.name.split(".")[1]
836
+ archive = dl_manager.download(config.data_url)
837
+ split_filenames = {
838
+ datasets.Split.TRAIN: "translated_train.tsv" if lang != "en" else "train.tsv",
839
+ datasets.Split.VALIDATION: "dev_2k.tsv",
840
+ datasets.Split.TEST: "test_2k.tsv",
841
+ }
842
+ return [
843
+ datasets.SplitGenerator(
844
+ name=split,
845
+ gen_kwargs={"filepath": dl_manager.iter_archive(archive), "filename": split_filenames[split]},
846
+ )
847
+ for split in split_filenames
848
+ ]
849
+
850
+ @staticmethod
851
+ def generate_examples(config=None, filepath=None, filename=None):
852
+ lang = config.name.split(".")[1]
853
+ for path, file in filepath:
854
+ if f"/{lang}/" in path and path.endswith(filename):
855
+ lines = (line.decode("utf-8") for line in file)
856
+ data = csv.reader(lines, delimiter="\t")
857
+ next(data) # skip header
858
+ for id_, row in enumerate(data):
859
+ if len(row) == 4:
860
+ yield id_, {
861
+ "sentence1": row[1],
862
+ "sentence2": row[2],
863
+ "label": row[3],
864
+ }
865
+
866
+
867
  class UdposParser:
868
 
869
  features = datasets.Features(