carmentano
commited on
Commit
•
ab7501f
1
Parent(s):
0fe1415
Upload wnli.py
Browse files
wnli.py
ADDED
@@ -0,0 +1,106 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Loading script for the TECA dataset.
|
2 |
+
import json
|
3 |
+
import datasets
|
4 |
+
|
5 |
+
logger = datasets.logging.get_logger(__name__)
|
6 |
+
|
7 |
+
_CITATION = """
|
8 |
+
ADD CITATION
|
9 |
+
"""
|
10 |
+
|
11 |
+
_DESCRIPTION = """
|
12 |
+
professional translation into Catalan of Winograd NLI dataset as published in GLUE Benchmark.
|
13 |
+
The Winograd NLI dataset presents 855 sentence pairs,
|
14 |
+
in which the first sentence contains an ambiguity and the second one a possible interpretation of it.
|
15 |
+
The label indicates if the interpretation is correct (1) or not (0).
|
16 |
+
"""
|
17 |
+
|
18 |
+
_HOMEPAGE = """https://cs.nyu.edu/~davise/papers/WinogradSchemas/WS.html"""
|
19 |
+
|
20 |
+
# TODO: upload datasets to github
|
21 |
+
_URL = "./"
|
22 |
+
_TRAINING_FILE = "wnli-train-ca.tsv"
|
23 |
+
_DEV_FILE = "wnli-dev-ca.tsv"
|
24 |
+
_TEST_FILE = "wnli-test-shuffled-ca.tsv"
|
25 |
+
|
26 |
+
|
27 |
+
class WinogradConfig(datasets.BuilderConfig):
|
28 |
+
""" Builder config for the Winograd-CA dataset """
|
29 |
+
|
30 |
+
def __init__(self, **kwargs):
|
31 |
+
"""BuilderConfig for Winograd-CA.
|
32 |
+
Args:
|
33 |
+
**kwargs: keyword arguments forwarded to super.
|
34 |
+
"""
|
35 |
+
super(WinogradConfig, self).__init__(**kwargs)
|
36 |
+
|
37 |
+
|
38 |
+
class Winograd(datasets.GeneratorBasedBuilder):
|
39 |
+
""" Winograd Dataset """
|
40 |
+
|
41 |
+
BUILDER_CONFIGS = [
|
42 |
+
WinogradConfig(
|
43 |
+
name="winograd",
|
44 |
+
version=datasets.Version("1.0.0"),
|
45 |
+
description="Winograd dataset",
|
46 |
+
),
|
47 |
+
]
|
48 |
+
|
49 |
+
def _info(self):
|
50 |
+
return datasets.DatasetInfo(
|
51 |
+
description=_DESCRIPTION,
|
52 |
+
features=datasets.Features(
|
53 |
+
{
|
54 |
+
"sentence1": datasets.Value("string"),
|
55 |
+
"sentence2": datasets.Value("string"),
|
56 |
+
"label": datasets.features.ClassLabel
|
57 |
+
(names=
|
58 |
+
[
|
59 |
+
"not_entailment",
|
60 |
+
"entailment"
|
61 |
+
]
|
62 |
+
),
|
63 |
+
}
|
64 |
+
),
|
65 |
+
homepage=_HOMEPAGE,
|
66 |
+
citation=_CITATION,
|
67 |
+
)
|
68 |
+
|
69 |
+
def _split_generators(self, dl_manager):
|
70 |
+
"""Returns SplitGenerators."""
|
71 |
+
urls_to_download = {
|
72 |
+
"train": f"{_URL}{_TRAINING_FILE}",
|
73 |
+
"dev": f"{_URL}{_DEV_FILE}",
|
74 |
+
"test": f"{_URL}{_TEST_FILE}",
|
75 |
+
}
|
76 |
+
downloaded_files = dl_manager.download_and_extract(urls_to_download)
|
77 |
+
|
78 |
+
return [
|
79 |
+
datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"]}),
|
80 |
+
datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"]}),
|
81 |
+
datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": downloaded_files["test"]}),
|
82 |
+
]
|
83 |
+
|
84 |
+
def _generate_examples(self, filepath):
|
85 |
+
"""This function returns the examples in the raw (text) form."""
|
86 |
+
logger.info("generating examples from = %s", filepath)
|
87 |
+
with open(filepath, encoding="utf-8") as f:
|
88 |
+
header = next(f)
|
89 |
+
process_label = {'0': "not_entailment", '1': "entailment"}
|
90 |
+
for id_, row in enumerate(f):
|
91 |
+
if "label" in header:
|
92 |
+
ref, sentence1, sentence2, score = row[:-1].split('\t')
|
93 |
+
yield id_, {
|
94 |
+
"sentence1": sentence1,
|
95 |
+
"sentence2": sentence2,
|
96 |
+
"label": process_label[score],
|
97 |
+
}
|
98 |
+
else:
|
99 |
+
ref, sentence1, sentence2 = row.split('\t')
|
100 |
+
yield id_, {
|
101 |
+
"sentence1": sentence1,
|
102 |
+
"sentence2": sentence2,
|
103 |
+
"label": -1,
|
104 |
+
}
|
105 |
+
|
106 |
+
|