BorisAlbar commited on
Commit
8993e9b
1 Parent(s): 94f9fd0

Dataset that yield squad_v2 compatible data

Browse files
Files changed (1) hide show
  1. frenchQA.py +115 -0
frenchQA.py ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """FrenchQA: One French QA Dataset to rule them all"""
2
+
3
+
4
+ import csv
5
+
6
+ import datasets
7
+ from datasets.tasks import QuestionAnsweringExtractive
8
+
9
+
10
+ # TODO(squad_v2): BibTeX citation
11
+ _CITATION = """\
12
+ """
13
+
14
+ _DESCRIPTION = """\
15
+ One French QA Dataset to rule them all, One French QA Dataset to find them, One French QA Dataset to bring them all, and in the darkness bind them.
16
+ """
17
+
18
+ _URLS = {
19
+ "train": "train.csv",
20
+ "dev": "valid.csv",
21
+ "test": "test.csv"
22
+ }
23
+
24
+
25
+ class FrenchQAConfig(datasets.BuilderConfig):
26
+ """BuilderConfig for frenchQA."""
27
+
28
+ def __init__(self, **kwargs):
29
+ """BuilderConfig for FrenchQA.
30
+ Args:
31
+ **kwargs: keyword arguments forwarded to super.
32
+ """
33
+ super(FrenchQAConfig, self).__init__(**kwargs)
34
+
35
+
36
+ class FrenchQA(datasets.GeneratorBasedBuilder):
37
+ """TODO(squad_v2): Short description of my dataset."""
38
+
39
+ # TODO(squad_v2): Set up version.
40
+ BUILDER_CONFIGS = [
41
+ FrenchQAConfig(name="frenchQA", version=datasets.Version("1.0.0"), description="frenchQA"),
42
+ ]
43
+
44
+ def _info(self):
45
+ # TODO(squad_v2): Specifies the datasets.DatasetInfo object
46
+ return datasets.DatasetInfo(
47
+ # This is the description that will appear on the datasets page.
48
+ description=_DESCRIPTION,
49
+ # datasets.features.FeatureConnectors
50
+ features=datasets.Features(
51
+ {
52
+ "id": datasets.Value("string"),
53
+ "title": datasets.Value("string"),
54
+ "context": datasets.Value("string"),
55
+ "question": datasets.Value("string"),
56
+ "answers": datasets.features.Sequence(
57
+ {
58
+ "text": datasets.Value("string"),
59
+ "answer_start": datasets.Value("int32"),
60
+ }
61
+ ),
62
+ # These are the features of your dataset like images, labels ...
63
+ }
64
+ ),
65
+ # If there's a common (input, target) tuple from the features,
66
+ # specify them here. They'll be used if as_supervised=True in
67
+ # builder.as_dataset.
68
+ supervised_keys=None,
69
+ # Homepage of the dataset for documentation
70
+ homepage="",
71
+ citation=_CITATION,
72
+ task_templates=[
73
+ QuestionAnsweringExtractive(
74
+ question_column="question", context_column="context", answers_column="answers"
75
+ )
76
+ ],
77
+ )
78
+
79
+ def _split_generators(self, dl_manager):
80
+ """Returns SplitGenerators."""
81
+ # TODO(squad_v2): Downloads the data and defines the splits
82
+ # dl_manager is a datasets.download.DownloadManager that can be used to
83
+ # download and extract URLs
84
+ urls_to_download = _URLS
85
+ downloaded_files = dl_manager.download_and_extract(urls_to_download)
86
+
87
+ return [
88
+ datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"]}),
89
+ datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"]}),
90
+ datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": downloaded_files["test"]}),
91
+ ]
92
+
93
+ def _generate_examples(self, filepath):
94
+ """Yields examples."""
95
+ # TODO(squad_v2): Yields (key, example) tuples from the dataset
96
+ with open(filepath, encoding="utf-8") as f:
97
+ squad = csv.DictReader(f, delimiter = ";")
98
+ for id_, row in enumerate(squad):
99
+ answer_start = []
100
+ text = []
101
+
102
+ if row["answer_start"] != "-1":
103
+ answer_start = [row["answer_start"]]
104
+ text = [row["answer"]]
105
+
106
+ yield id_, {
107
+ "title": row["dataset"],
108
+ "context": row["context"],
109
+ "question": row["question"],
110
+ "id": id_,
111
+ "answers": {
112
+ "answer_start": answer_start,
113
+ "text": text,
114
+ },
115
+ }