sileod commited on
Commit
d45e461
·
1 Parent(s): 152bf92

Create recast.py

Browse files
Files changed (1) hide show
  1. recast.py +230 -0
recast.py ADDED
@@ -0,0 +1,230 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ # Lint as: python3
17
+ """Recast datasets"""
18
+
19
+ from __future__ import absolute_import, division, print_function
20
+
21
+ import csv
22
+ import os
23
+ import textwrap
24
+
25
+ import six
26
+
27
+ import datasets
28
+
29
+
30
+ _Recast_CITATION = r"""inproceedings{poliak-etal-2018-collecting,
31
+ title = "Collecting Diverse Natural Language Inference Problems for Sentence Representation Evaluation",
32
+ author = "Poliak, Adam and
33
+ Haldar, Aparajita and
34
+ Rudinger, Rachel and
35
+ Hu, J. Edward and
36
+ Pavlick, Ellie and
37
+ White, Aaron Steven and
38
+ Van Durme, Benjamin",
39
+ booktitle = "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
40
+ month = oct # "-" # nov,
41
+ year = "2018",
42
+ address = "Brussels, Belgium",
43
+ publisher = "Association for Computational Linguistics",
44
+ url = "https://aclanthology.org/D18-1007",
45
+ doi = "10.18653/v1/D18-1007",
46
+ pages = "67--81",
47
+ abstract = "We present a large-scale collection of diverse natural language inference (NLI) datasets that help provide insight into how well a sentence representation captures distinct types of reasoning. The collection results from recasting 13 existing datasets from 7 semantic phenomena into a common NLI structure, resulting in over half a million labeled context-hypothesis pairs in total. We refer to our collection as the DNC: Diverse Natural Language Inference Collection. The DNC is available online at \url{https://www.decomp.net}, and will grow over time as additional resources are recast and added from novel sources.",
48
+ }
49
+ """
50
+
51
+ _Recast_DESCRIPTION = """\
52
+ A diverse collection of tasks recasted as natural language inference tasks.
53
+ """
54
+
55
+ DATA_URL = "https://www.dropbox.com/s/z1mcq6ygfsae0wj/recast.zip?dl=1"
56
+
57
+ TASK_TO_LABELS = {
58
+ "recast_kg_relations": ["1", "2", "3", "4", "5", "6"],
59
+ "recast_puns": ["not-entailed", "entailed"],
60
+ "recast_factuality": ["not-entailed", "entailed"],
61
+ "recast_verbnet": ["not-entailed", "entailed"],
62
+ "recast_verbcorner": ["not-entailed", "entailed"],
63
+ "recast_sentiment": ["not-entailed", "entailed"],
64
+ "recast_megaveridicality": ["not-entailed", "entailed"],
65
+ "recast_ner": ["not-entailed", "entailed"],
66
+ "recast_winogender": ["not-entailed", "entailed"],
67
+ "recast_ner": ["not-entailed", "entailed"],
68
+ }
69
+
70
+
71
+ def get_labels(task):
72
+ return TASK_TO_LABELS[task]
73
+
74
+
75
+ class RecastConfig(datasets.BuilderConfig):
76
+ """BuilderConfig for Recast."""
77
+
78
+ def __init__(
79
+ self,
80
+ text_features,
81
+ label_classes=None,
82
+ process_label=lambda x: x,
83
+ **kwargs,
84
+ ):
85
+ """BuilderConfig for Recast.
86
+ Args:
87
+ text_features: `dict[string, string]`, map from the name of the feature
88
+ dict for each text field to the name of the column in the tsv file
89
+ label_column: `string`, name of the column in the tsv file corresponding
90
+ to the label
91
+ data_url: `string`, url to download the zip file from
92
+ data_dir: `string`, the path to the folder containing the tsv files in the
93
+ downloaded zip
94
+ citation: `string`, citation for the data set
95
+ url: `string`, url for information about the data set
96
+ label_classes: `list[string]`, the list of classes if the label is
97
+ categorical. If not provided, then the label will be of type
98
+ `datasets.Value('float32')`.
99
+ process_label: `Function[string, any]`, function taking in the raw value
100
+ of the label and processing it to the form required by the label feature
101
+ **kwargs: keyword arguments forwarded to super.
102
+ """
103
+
104
+ super(RecastConfig, self).__init__(
105
+ version=datasets.Version("1.0.0", ""), **kwargs
106
+ )
107
+
108
+ self.text_features = text_features
109
+ self.label_column = "label"
110
+ self.label_classes = get_labels(self.name)
111
+ self.data_url = DATA_URL
112
+ self.data_dir = os.path.join("recast", self.name)
113
+ self.citation = textwrap.dedent(_Recast_CITATION)
114
+ self.process_label = lambda x: str(x)
115
+ self.description = ""
116
+ self.url = ""
117
+
118
+
119
+ class Recast(datasets.GeneratorBasedBuilder):
120
+
121
+ """The General Language Understanding Evaluation (Recast) benchmark."""
122
+
123
+ BUILDER_CONFIG_CLASS = RecastConfig
124
+
125
+ BUILDER_CONFIGS = [
126
+ RecastConfig(
127
+ name="recast_kg_relations",
128
+ text_features={"context": "context", "hypothesis": "hypothesis"},
129
+ ),
130
+ RecastConfig(
131
+ name="recast_puns",
132
+ text_features={"context": "context", "hypothesis": "hypothesis"},
133
+ ),
134
+ RecastConfig(
135
+ name="recast_factuality",
136
+ text_features={"context": "context", "hypothesis": "hypothesis"},
137
+ ),
138
+ RecastConfig(
139
+ name="recast_verbnet",
140
+ text_features={"context": "context", "hypothesis": "hypothesis"},
141
+ ),
142
+ RecastConfig(
143
+ name="recast_verbcorner",
144
+ text_features={"context": "context", "hypothesis": "hypothesis"},
145
+ ),
146
+ RecastConfig(
147
+ name="recast_ner",
148
+ text_features={"context": "context", "hypothesis": "hypothesis"},
149
+ ),
150
+ RecastConfig(
151
+ name="recast_sentiment",
152
+ text_features={"context": "context", "hypothesis": "hypothesis"},
153
+ ),
154
+ RecastConfig(
155
+ name="recast_megaveridicality",
156
+ text_features={"context": "context", "hypothesis": "hypothesis"},
157
+ ),
158
+ ]
159
+
160
+ def _info(self):
161
+ features = {
162
+ text_feature: datasets.Value("string")
163
+ for text_feature in six.iterkeys(self.config.text_features)
164
+ }
165
+ if self.config.label_classes:
166
+ features["label"] = datasets.features.ClassLabel(
167
+ names=self.config.label_classes
168
+ )
169
+ else:
170
+ features["label"] = datasets.Value("float32")
171
+ features["idx"] = datasets.Value("int32")
172
+ return datasets.DatasetInfo(
173
+ description=_Recast_DESCRIPTION,
174
+ features=datasets.Features(features),
175
+ homepage=self.config.url,
176
+ citation=self.config.citation + "\n" + _Recast_CITATION,
177
+ )
178
+
179
+ def _split_generators(self, dl_manager):
180
+ dl_dir = dl_manager.download_and_extract(self.config.data_url)
181
+ data_dir = os.path.join(dl_dir, self.config.data_dir)
182
+
183
+ return [
184
+ datasets.SplitGenerator(
185
+ name=datasets.Split.TRAIN,
186
+ gen_kwargs={
187
+ "data_file": os.path.join(data_dir or "", "train.tsv"),
188
+ "split": "train",
189
+ },
190
+ ),
191
+ datasets.SplitGenerator(
192
+ name=datasets.Split.VALIDATION,
193
+ gen_kwargs={
194
+ "data_file": os.path.join(data_dir or "", "dev.tsv"),
195
+ "split": "dev",
196
+ },
197
+ ),
198
+ datasets.SplitGenerator(
199
+ name=datasets.Split.TEST,
200
+ gen_kwargs={
201
+ "data_file": os.path.join(data_dir or "", "test.tsv"),
202
+ "split": "test",
203
+ },
204
+ ),
205
+ ]
206
+
207
+ def _generate_examples(self, data_file, split):
208
+
209
+ process_label = self.config.process_label
210
+ label_classes = self.config.label_classes
211
+
212
+ with open(data_file, encoding="utf8") as f:
213
+ reader = csv.DictReader(f, delimiter="\t", quoting=csv.QUOTE_NONE)
214
+
215
+ for n, row in enumerate(reader):
216
+
217
+ example = {
218
+ feat: row[col]
219
+ for feat, col in six.iteritems(self.config.text_features)
220
+ }
221
+ example["idx"] = n
222
+
223
+ if self.config.label_column in row:
224
+ label = row[self.config.label_column]
225
+ if label_classes and label not in label_classes:
226
+ label = int(label) if label else None
227
+ example["label"] = process_label(label)
228
+ else:
229
+ example["label"] = process_label(-1)
230
+ yield example["idx"], example