Datasets:
Robin Kurtz
commited on
Commit
·
c9ac0a9
1
Parent(s):
9477631
processing script
Browse files- overlim.py +483 -0
overlim.py
ADDED
@@ -0,0 +1,483 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 |
+
"""The SuperGLUE benchmark."""
|
18 |
+
|
19 |
+
import json
|
20 |
+
import os
|
21 |
+
|
22 |
+
import datasets
|
23 |
+
|
24 |
+
_CITATION = """\
|
25 |
+
"""
|
26 |
+
|
27 |
+
# You can copy an official description
|
28 |
+
_DESCRIPTION = """\
|
29 |
+
"""
|
30 |
+
|
31 |
+
_HOMEPAGE = ""
|
32 |
+
|
33 |
+
_LICENSE = ""
|
34 |
+
|
35 |
+
_GLUE_CITATION = """\
|
36 |
+
@inproceedings{wang2019glue,
|
37 |
+
title={{GLUE}: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding},
|
38 |
+
author={Wang, Alex and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R.},
|
39 |
+
note={In the Proceedings of ICLR.},
|
40 |
+
year={2019}
|
41 |
+
}
|
42 |
+
"""
|
43 |
+
|
44 |
+
_GLUE_DESCRIPTION = """\
|
45 |
+
GLUE, the General Language Understanding Evaluation benchmark
|
46 |
+
(https://gluebenchmark.com/) is a collection of resources for training,
|
47 |
+
evaluating, and analyzing natural language understanding systems.
|
48 |
+
|
49 |
+
"""
|
50 |
+
_SST_DESCRIPTION = """\
|
51 |
+
The Stanford Sentiment Treebank consists of sentences from movie reviews and
|
52 |
+
human annotations of their sentiment. The task is to predict the sentiment of a
|
53 |
+
given sentence. We use the two-way (positive/negative) class split, and use only
|
54 |
+
sentence-level labels."""
|
55 |
+
_SST_CITATION = """\
|
56 |
+
@inproceedings{socher2013recursive,
|
57 |
+
title={Recursive deep models for semantic compositionality over a sentiment treebank},
|
58 |
+
author={Socher, Richard and Perelygin, Alex and Wu, Jean and Chuang, Jason and Manning, Christopher D and Ng, Andrew and Potts, Christopher},
|
59 |
+
booktitle={Proceedings of the 2013 conference on empirical methods in natural language processing},
|
60 |
+
pages={1631--1642},
|
61 |
+
year={2013}
|
62 |
+
}"""
|
63 |
+
_MRPC_DESCRIPTION = """\
|
64 |
+
The Microsoft Research Paraphrase Corpus (Dolan & Brockett, 2005) is a corpus of
|
65 |
+
sentence pairs automatically extracted from online news sources, with human annotations
|
66 |
+
for whether the sentences in the pair are semantically equivalent."""
|
67 |
+
_MRPC_CITATION = """\
|
68 |
+
@inproceedings{dolan2005automatically,
|
69 |
+
title={Automatically constructing a corpus of sentential paraphrases},
|
70 |
+
author={Dolan, William B and Brockett, Chris},
|
71 |
+
booktitle={Proceedings of the Third International Workshop on Paraphrasing (IWP2005)},
|
72 |
+
year={2005}
|
73 |
+
}"""
|
74 |
+
_QQP_DESCRIPTION = """\
|
75 |
+
The Quora Question Pairs2 dataset is a collection of question pairs from the
|
76 |
+
community question-answering website Quora. The task is to determine whether a
|
77 |
+
pair of questions are semantically equivalent."""
|
78 |
+
_QQP_CITATION = """\
|
79 |
+
@online{WinNT,
|
80 |
+
author = {Iyer, Shankar and Dandekar, Nikhil and Csernai, Kornel},
|
81 |
+
title = {First Quora Dataset Release: Question Pairs},
|
82 |
+
year = {2017},
|
83 |
+
url = {https://data.quora.com/First-Quora-Dataset-Release-Question-Pairs},
|
84 |
+
urldate = {2019-04-03}
|
85 |
+
}"""
|
86 |
+
_STSB_DESCRIPTION = """\
|
87 |
+
The Semantic Textual Similarity Benchmark (Cer et al., 2017) is a collection of
|
88 |
+
sentence pairs drawn from news headlines, video and image captions, and natural
|
89 |
+
language inference data. Each pair is human-annotated with a similarity score
|
90 |
+
from 1 to 5."""
|
91 |
+
_STSB_CITATION = """\
|
92 |
+
@article{cer2017semeval,
|
93 |
+
title={Semeval-2017 task 1: Semantic textual similarity-multilingual and cross-lingual focused evaluation},
|
94 |
+
author={Cer, Daniel and Diab, Mona and Agirre, Eneko and Lopez-Gazpio, Inigo and Specia, Lucia},
|
95 |
+
journal={arXiv preprint arXiv:1708.00055},
|
96 |
+
year={2017}
|
97 |
+
}"""
|
98 |
+
_MNLI_DESCRIPTION = """\
|
99 |
+
The Multi-Genre Natural Language Inference Corpus is a crowdsourced
|
100 |
+
collection of sentence pairs with textual entailment annotations. Given a premise sentence
|
101 |
+
and a hypothesis sentence, the task is to predict whether the premise entails the hypothesis
|
102 |
+
(entailment), contradicts the hypothesis (contradiction), or neither (neutral). The premise sentences are
|
103 |
+
gathered from ten different sources, including transcribed speech, fiction, and government reports.
|
104 |
+
We use the standard test set, for which we obtained private labels from the authors, and evaluate
|
105 |
+
on both the matched (in-domain) and mismatched (cross-domain) section. We also use and recommend
|
106 |
+
the SNLI corpus as 550k examples of auxiliary training data."""
|
107 |
+
_MNLI_CITATION = """\
|
108 |
+
@InProceedings{N18-1101,
|
109 |
+
author = "Williams, Adina
|
110 |
+
and Nangia, Nikita
|
111 |
+
and Bowman, Samuel",
|
112 |
+
title = "A Broad-Coverage Challenge Corpus for
|
113 |
+
Sentence Understanding through Inference",
|
114 |
+
booktitle = "Proceedings of the 2018 Conference of
|
115 |
+
the North American Chapter of the
|
116 |
+
Association for Computational Linguistics:
|
117 |
+
Human Language Technologies, Volume 1 (Long
|
118 |
+
Papers)",
|
119 |
+
year = "2018",
|
120 |
+
publisher = "Association for Computational Linguistics",
|
121 |
+
pages = "1112--1122",
|
122 |
+
location = "New Orleans, Louisiana",
|
123 |
+
url = "http://aclweb.org/anthology/N18-1101"
|
124 |
+
}
|
125 |
+
@article{bowman2015large,
|
126 |
+
title={A large annotated corpus for learning natural language inference},
|
127 |
+
author={Bowman, Samuel R and Angeli, Gabor and Potts, Christopher and Manning, Christopher D},
|
128 |
+
journal={arXiv preprint arXiv:1508.05326},
|
129 |
+
year={2015}
|
130 |
+
}"""
|
131 |
+
_QNLI_DESCRIPTION = """\
|
132 |
+
The Stanford Question Answering Dataset is a question-answering
|
133 |
+
dataset consisting of question-paragraph pairs, where one of the sentences in the paragraph (drawn
|
134 |
+
from Wikipedia) contains the answer to the corresponding question (written by an annotator). We
|
135 |
+
convert the task into sentence pair classification by forming a pair between each question and each
|
136 |
+
sentence in the corresponding context, and filtering out pairs with low lexical overlap between the
|
137 |
+
question and the context sentence. The task is to determine whether the context sentence contains
|
138 |
+
the answer to the question. This modified version of the original task removes the requirement that
|
139 |
+
the model select the exact answer, but also removes the simplifying assumptions that the answer
|
140 |
+
is always present in the input and that lexical overlap is a reliable cue."""
|
141 |
+
_QNLI_CITATION = """\
|
142 |
+
@article{rajpurkar2016squad,
|
143 |
+
title={Squad: 100,000+ questions for machine comprehension of text},
|
144 |
+
author={Rajpurkar, Pranav and Zhang, Jian and Lopyrev, Konstantin and Liang, Percy},
|
145 |
+
journal={arXiv preprint arXiv:1606.05250},
|
146 |
+
year={2016}
|
147 |
+
}"""
|
148 |
+
_WNLI_DESCRIPTION = """\
|
149 |
+
The Winograd Schema Challenge (Levesque et al., 2011) is a reading comprehension task
|
150 |
+
in which a system must read a sentence with a pronoun and select the referent of that pronoun from
|
151 |
+
a list of choices. The examples are manually constructed to foil simple statistical methods: Each
|
152 |
+
one is contingent on contextual information provided by a single word or phrase in the sentence.
|
153 |
+
To convert the problem into sentence pair classification, we construct sentence pairs by replacing
|
154 |
+
the ambiguous pronoun with each possible referent. The task is to predict if the sentence with the
|
155 |
+
pronoun substituted is entailed by the original sentence. We use a small evaluation set consisting of
|
156 |
+
new examples derived from fiction books that was shared privately by the authors of the original
|
157 |
+
corpus. While the included training set is balanced between two classes, the test set is imbalanced
|
158 |
+
between them (65% not entailment). Also, due to a data quirk, the development set is adversarial:
|
159 |
+
hypotheses are sometimes shared between training and development examples, so if a model memorizes the
|
160 |
+
training examples, they will predict the wrong label on corresponding development set
|
161 |
+
example. As with QNLI, each example is evaluated separately, so there is not a systematic correspondence
|
162 |
+
between a model's score on this task and its score on the unconverted original task. We
|
163 |
+
call converted dataset WNLI (Winograd NLI)."""
|
164 |
+
_WNLI_CITATION = """\
|
165 |
+
@inproceedings{levesque2012winograd,
|
166 |
+
title={The winograd schema challenge},
|
167 |
+
author={Levesque, Hector and Davis, Ernest and Morgenstern, Leora},
|
168 |
+
booktitle={Thirteenth International Conference on the Principles of Knowledge Representation and Reasoning},
|
169 |
+
year={2012}
|
170 |
+
}"""
|
171 |
+
|
172 |
+
_SUPER_GLUE_CITATION = """\
|
173 |
+
@article{wang2019superglue,
|
174 |
+
title={SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems},
|
175 |
+
author={Wang, Alex and Pruksachatkun, Yada and Nangia, Nikita and Singh, Amanpreet and Michael, Julian and Hill, Felix and Levy, Omer and Bowman, Samuel R},
|
176 |
+
journal={arXiv preprint arXiv:1905.00537},
|
177 |
+
year={2019}
|
178 |
+
}
|
179 |
+
|
180 |
+
Note that each SuperGLUE dataset has its own citation. Please see the source to
|
181 |
+
get the correct citation for each contained dataset.
|
182 |
+
"""
|
183 |
+
|
184 |
+
_SUPER_GLUE_DESCRIPTION = """\
|
185 |
+
SuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after
|
186 |
+
GLUE with a new set of more difficult language understanding tasks, improved
|
187 |
+
resources, and a new public leaderboard.
|
188 |
+
|
189 |
+
"""
|
190 |
+
|
191 |
+
_BOOLQ_DESCRIPTION = """\
|
192 |
+
BoolQ (Boolean Questions, Clark et al., 2019a) is a QA task where each example consists of a short
|
193 |
+
passage and a yes/no question about the passage. The questions are provided anonymously and
|
194 |
+
unsolicited by users of the Google search engine, and afterwards paired with a paragraph from a
|
195 |
+
Wikipedia article containing the answer. Following the original work, we evaluate with accuracy."""
|
196 |
+
|
197 |
+
_CB_DESCRIPTION = """\
|
198 |
+
The CommitmentBank (De Marneffe et al., 2019) is a corpus of short texts in which at least
|
199 |
+
one sentence contains an embedded clause. Each of these embedded clauses is annotated with the
|
200 |
+
degree to which we expect that the person who wrote the text is committed to the truth of the clause.
|
201 |
+
The resulting task framed as three-class textual entailment on examples that are drawn from the Wall
|
202 |
+
Street Journal, fiction from the British National Corpus, and Switchboard. Each example consists
|
203 |
+
of a premise containing an embedded clause and the corresponding hypothesis is the extraction of
|
204 |
+
that clause. We use a subset of the data that had inter-annotator agreement above 0.85. The data is
|
205 |
+
imbalanced (relatively fewer neutral examples), so we evaluate using accuracy and F1, where for
|
206 |
+
multi-class F1 we compute the unweighted average of the F1 per class."""
|
207 |
+
|
208 |
+
_COPA_DESCRIPTION = """\
|
209 |
+
The Choice Of Plausible Alternatives (COPA, Roemmele et al., 2011) dataset is a causal
|
210 |
+
reasoning task in which a system is given a premise sentence and two possible alternatives. The
|
211 |
+
system must choose the alternative which has the more plausible causal relationship with the premise.
|
212 |
+
The method used for the construction of the alternatives ensures that the task requires causal reasoning
|
213 |
+
to solve. Examples either deal with alternative possible causes or alternative possible effects of the
|
214 |
+
premise sentence, accompanied by a simple question disambiguating between the two instance
|
215 |
+
types for the model. All examples are handcrafted and focus on topics from online blogs and a
|
216 |
+
photography-related encyclopedia. Following the recommendation of the authors, we evaluate using
|
217 |
+
accuracy."""
|
218 |
+
|
219 |
+
_RTE_DESCRIPTION = """\
|
220 |
+
The Recognizing Textual Entailment (RTE) datasets come from a series of annual competitions
|
221 |
+
on textual entailment, the problem of predicting whether a given premise sentence entails a given
|
222 |
+
hypothesis sentence (also known as natural language inference, NLI). RTE was previously included
|
223 |
+
in GLUE, and we use the same data and format as before: We merge data from RTE1 (Dagan
|
224 |
+
et al., 2006), RTE2 (Bar Haim et al., 2006), RTE3 (Giampiccolo et al., 2007), and RTE5 (Bentivogli
|
225 |
+
et al., 2009). All datasets are combined and converted to two-class classification: entailment and
|
226 |
+
not_entailment. Of all the GLUE tasks, RTE was among those that benefited from transfer learning
|
227 |
+
the most, jumping from near random-chance performance (~56%) at the time of GLUE's launch to
|
228 |
+
85% accuracy (Liu et al., 2019c) at the time of writing. Given the eight point gap with respect to
|
229 |
+
human performance, however, the task is not yet solved by machines, and we expect the remaining
|
230 |
+
gap to be difficult to close."""
|
231 |
+
|
232 |
+
_BOOLQ_CITATION = """\
|
233 |
+
@inproceedings{clark2019boolq,
|
234 |
+
title={BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions},
|
235 |
+
author={Clark, Christopher and Lee, Kenton and Chang, Ming-Wei, and Kwiatkowski, Tom and Collins, Michael, and Toutanova, Kristina},
|
236 |
+
booktitle={NAACL},
|
237 |
+
year={2019}
|
238 |
+
}"""
|
239 |
+
|
240 |
+
_CB_CITATION = """\
|
241 |
+
@article{de marneff_simons_tonhauser_2019,
|
242 |
+
title={The CommitmentBank: Investigating projection in naturally occurring discourse},
|
243 |
+
journal={proceedings of Sinn und Bedeutung 23},
|
244 |
+
author={De Marneff, Marie-Catherine and Simons, Mandy and Tonhauser, Judith},
|
245 |
+
year={2019}
|
246 |
+
}"""
|
247 |
+
|
248 |
+
_COPA_CITATION = """\
|
249 |
+
@inproceedings{roemmele2011choice,
|
250 |
+
title={Choice of plausible alternatives: An evaluation of commonsense causal reasoning},
|
251 |
+
author={Roemmele, Melissa and Bejan, Cosmin Adrian and Gordon, Andrew S},
|
252 |
+
booktitle={2011 AAAI Spring Symposium Series},
|
253 |
+
year={2011}
|
254 |
+
}"""
|
255 |
+
|
256 |
+
_RTE_CITATION = """\
|
257 |
+
@inproceedings{dagan2005pascal,
|
258 |
+
title={The PASCAL recognising textual entailment challenge},
|
259 |
+
author={Dagan, Ido and Glickman, Oren and Magnini, Bernardo},
|
260 |
+
booktitle={Machine Learning Challenges Workshop},
|
261 |
+
pages={177--190},
|
262 |
+
year={2005},
|
263 |
+
organization={Springer}
|
264 |
+
}
|
265 |
+
@inproceedings{bar2006second,
|
266 |
+
title={The second pascal recognising textual entailment challenge},
|
267 |
+
author={Bar-Haim, Roy and Dagan, Ido and Dolan, Bill and Ferro, Lisa and Giampiccolo, Danilo and Magnini, Bernardo and Szpektor, Idan},
|
268 |
+
booktitle={Proceedings of the second PASCAL challenges workshop on recognising textual entailment},
|
269 |
+
volume={6},
|
270 |
+
number={1},
|
271 |
+
pages={6--4},
|
272 |
+
year={2006},
|
273 |
+
organization={Venice}
|
274 |
+
}
|
275 |
+
@inproceedings{giampiccolo2007third,
|
276 |
+
title={The third pascal recognizing textual entailment challenge},
|
277 |
+
author={Giampiccolo, Danilo and Magnini, Bernardo and Dagan, Ido and Dolan, Bill},
|
278 |
+
booktitle={Proceedings of the ACL-PASCAL workshop on textual entailment and paraphrasing},
|
279 |
+
pages={1--9},
|
280 |
+
year={2007},
|
281 |
+
organization={Association for Computational Linguistics}
|
282 |
+
}
|
283 |
+
@inproceedings{bentivogli2009fifth,
|
284 |
+
title={The Fifth PASCAL Recognizing Textual Entailment Challenge.},
|
285 |
+
author={Bentivogli, Luisa and Clark, Peter and Dagan, Ido and Giampiccolo, Danilo},
|
286 |
+
booktitle={TAC},
|
287 |
+
year={2009}
|
288 |
+
}"""
|
289 |
+
|
290 |
+
# TODO: Add link to the official dataset URLs here
|
291 |
+
# The HuggingFace Datasets library doesn't host the datasets but only points to the original files.
|
292 |
+
# This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
|
293 |
+
_URL = "https://huggingface.co/datasets/KBLab/overlim/resolve/main/data/"
|
294 |
+
_TASKS = {
|
295 |
+
"boolq": "boolq.tar.gz",
|
296 |
+
"cb": "cb.tar.gz",
|
297 |
+
"copa": "copa.tar.gz",
|
298 |
+
"mnli": "mnli.tar.gz",
|
299 |
+
"mrpc": "mrpc.tar.gz",
|
300 |
+
"qnli": "qnli.tar.gz",
|
301 |
+
"qqp": "qqp.tar.gz",
|
302 |
+
"rte": "rte.tar.gz",
|
303 |
+
"sst": "sst.tar.gz",
|
304 |
+
"stsb": "stsb.tar.gz",
|
305 |
+
"wnli": "wnli.tar.gz"
|
306 |
+
}
|
307 |
+
_LANGUAGES = {"sv", "da", "nb"}
|
308 |
+
|
309 |
+
|
310 |
+
class OverLimConfig(datasets.BuilderConfig):
|
311 |
+
"""BuilderConfig for Suc."""
|
312 |
+
def __init__(self, name, description, features, citation, language, label_classes=("False", "True"), **kwargs):
|
313 |
+
"""BuilderConfig for OverLim.
|
314 |
+
"""
|
315 |
+
self.full_name = name + "_" + language
|
316 |
+
super(OverLimConfig,
|
317 |
+
self).__init__(name=self.full_name , version=datasets.Version("1.0.2"), **kwargs)
|
318 |
+
self.features = features
|
319 |
+
self.label_classes = label_classes
|
320 |
+
self.citation = citation
|
321 |
+
self.description = description
|
322 |
+
# self.name = name
|
323 |
+
self.language = language
|
324 |
+
|
325 |
+
|
326 |
+
|
327 |
+
class OverLim(datasets.GeneratorBasedBuilder):
|
328 |
+
"""OverLim"""
|
329 |
+
|
330 |
+
BUILDER_CONFIGS = [[OverLimConfig(
|
331 |
+
name="boolq",
|
332 |
+
description=_BOOLQ_DESCRIPTION,
|
333 |
+
features=["question", "passage"],
|
334 |
+
label_classes=["False", "True"],
|
335 |
+
citation=_BOOLQ_CITATION,
|
336 |
+
language=lang,
|
337 |
+
),
|
338 |
+
OverLimConfig(
|
339 |
+
name="cb",
|
340 |
+
description=_CB_DESCRIPTION,
|
341 |
+
features=["premise", "hypothesis"],
|
342 |
+
label_classes=["entailment", "contradiction", "neutral"],
|
343 |
+
citation=_CB_CITATION,
|
344 |
+
language=lang,
|
345 |
+
),
|
346 |
+
OverLimConfig(
|
347 |
+
name="copa",
|
348 |
+
description=_COPA_DESCRIPTION,
|
349 |
+
label_classes=["choice1", "choice2"],
|
350 |
+
# Note that question will only be the X in the statement "What's
|
351 |
+
# the X for this?".
|
352 |
+
features=["premise", "choice1", "choice2", "question"],
|
353 |
+
citation=_COPA_CITATION,
|
354 |
+
language=lang,
|
355 |
+
),
|
356 |
+
OverLimConfig(
|
357 |
+
name="rte",
|
358 |
+
description=_RTE_DESCRIPTION,
|
359 |
+
features=["premise", "hypothesis"],
|
360 |
+
label_classes=["entailment", "not_entailment"],
|
361 |
+
citation=_RTE_CITATION,
|
362 |
+
language=lang,
|
363 |
+
),
|
364 |
+
OverLimConfig(
|
365 |
+
name="qqp",
|
366 |
+
description=_QQP_DESCRIPTION,
|
367 |
+
features=["text_a", "text_b"],
|
368 |
+
label_classes=["not_duplicate", "duplicate"],
|
369 |
+
citation=_QQP_CITATION,
|
370 |
+
language=lang,
|
371 |
+
),
|
372 |
+
OverLimConfig(
|
373 |
+
name="qnli",
|
374 |
+
description=_QNLI_DESCRIPTION,
|
375 |
+
features=["premise", "hypothesis"],
|
376 |
+
label_classes=["entailment", "not_entailment"],
|
377 |
+
citation=_QNLI_CITATION,
|
378 |
+
language=lang,
|
379 |
+
),
|
380 |
+
OverLimConfig(
|
381 |
+
name="stsb",
|
382 |
+
description=_STSB_DESCRIPTION,
|
383 |
+
features=["text_a", "text_b"],
|
384 |
+
citation=_STSB_CITATION,
|
385 |
+
language=lang,
|
386 |
+
),
|
387 |
+
OverLimConfig(
|
388 |
+
name="mnli",
|
389 |
+
description=_MNLI_DESCRIPTION,
|
390 |
+
features=["premise", "hypothesis"],
|
391 |
+
label_classes=["entailment", "neutral", "contradiction"],
|
392 |
+
citation=_MNLI_CITATION,
|
393 |
+
language=lang,
|
394 |
+
),
|
395 |
+
OverLimConfig(
|
396 |
+
name="mrpc",
|
397 |
+
description=_MRPC_DESCRIPTION,
|
398 |
+
features=["text_a", "text_b"],
|
399 |
+
label_classes=["not_equivalent", "equivalent"],
|
400 |
+
citation=_MRPC_CITATION,
|
401 |
+
language=lang,
|
402 |
+
),
|
403 |
+
OverLimConfig(
|
404 |
+
name="wnli",
|
405 |
+
description=_WNLI_DESCRIPTION,
|
406 |
+
features=["premise", "hypothesis"],
|
407 |
+
label_classes=["not_entailment", "entailment"],
|
408 |
+
citation=_WNLI_CITATION,
|
409 |
+
language=lang,
|
410 |
+
),
|
411 |
+
OverLimConfig(
|
412 |
+
name="sst",
|
413 |
+
description=_SST_DESCRIPTION,
|
414 |
+
features=["text"],
|
415 |
+
label_classes=["negative", "positive"],
|
416 |
+
citation=_SST_CITATION,
|
417 |
+
language=lang,
|
418 |
+
)
|
419 |
+
|
420 |
+
] for lang in _LANGUAGES]
|
421 |
+
BUILDER_CONFIGS = [element for inner in BUILDER_CONFIGS for element in inner]
|
422 |
+
|
423 |
+
def _info(self):
|
424 |
+
features = {feature: datasets.Value("string") for feature in self.config.features}
|
425 |
+
features["idx"] = datasets.Value("int32")
|
426 |
+
|
427 |
+
return datasets.DatasetInfo(
|
428 |
+
description=_GLUE_DESCRIPTION + self.config.description,
|
429 |
+
features=datasets.Features(features),
|
430 |
+
homepage=_HOMEPAGE,
|
431 |
+
citation=self.config.citation + "\n" + _SUPER_GLUE_CITATION,
|
432 |
+
)
|
433 |
+
|
434 |
+
def _split_generators(self, dl_manager):
|
435 |
+
dl_dir = dl_manager.download_and_extract(os.path.join(_URL, self.config.lang, self.config.data_url))
|
436 |
+
dl_dir = os.path.join(_URL, self.config.lang, self.config.name)
|
437 |
+
return [
|
438 |
+
datasets.SplitGenerator(
|
439 |
+
name=datasets.Split.TRAIN,
|
440 |
+
gen_kwargs={
|
441 |
+
"data_file": os.path.join(dl_dir, "train.jsonl"),
|
442 |
+
},
|
443 |
+
),
|
444 |
+
datasets.SplitGenerator(
|
445 |
+
name=datasets.Split.VALIDATION,
|
446 |
+
gen_kwargs={
|
447 |
+
"data_file": os.path.join(dl_dir, "val.jsonl"),
|
448 |
+
},
|
449 |
+
),
|
450 |
+
datasets.SplitGenerator(
|
451 |
+
name=datasets.Split.TEST,
|
452 |
+
gen_kwargs={
|
453 |
+
"data_file": os.path.join(dl_dir, "test.jsonl"),
|
454 |
+
},
|
455 |
+
),
|
456 |
+
]
|
457 |
+
|
458 |
+
def _generate_examples(self, data_file):
|
459 |
+
with open(data_file, encoding="utf-8") as f:
|
460 |
+
for line in f:
|
461 |
+
row = json.loads(line)
|
462 |
+
example = {feature: row[feature] for feature in self.config.features}
|
463 |
+
example["idx"] = row["idx"]
|
464 |
+
|
465 |
+
if self.config.name == "copa":
|
466 |
+
example["label"] = "choice2" if row["label"] else "choice1"
|
467 |
+
else:
|
468 |
+
example["label"] = _cast_label(row["label"])
|
469 |
+
yield example["idx"], example
|
470 |
+
|
471 |
+
|
472 |
+
def _cast_label(label):
|
473 |
+
"""Converts the label into the appropriate string version."""
|
474 |
+
if isinstance(label, str):
|
475 |
+
return label
|
476 |
+
elif isinstance(label, bool):
|
477 |
+
return "True" if label else "False"
|
478 |
+
return label
|
479 |
+
# elif isinstance(label, int):
|
480 |
+
# assert label in (0, 1)
|
481 |
+
# return str(label)
|
482 |
+
# else:
|
483 |
+
# raise ValueError("Invalid label format.")
|