Commit
•
5dfd01a
1
Parent(s):
7510be0
Delete loading script
Browse files- scitail.py +0 -298
scitail.py
DELETED
@@ -1,298 +0,0 @@
|
|
1 |
-
"""TODO(sciTail): Add a description here."""
|
2 |
-
|
3 |
-
|
4 |
-
import csv
|
5 |
-
import json
|
6 |
-
import os
|
7 |
-
import textwrap
|
8 |
-
|
9 |
-
import datasets
|
10 |
-
|
11 |
-
|
12 |
-
# TODO(sciTail): BibTeX citation
|
13 |
-
_CITATION = """\
|
14 |
-
inproceedings{scitail,
|
15 |
-
Author = {Tushar Khot and Ashish Sabharwal and Peter Clark},
|
16 |
-
Booktitle = {AAAI},
|
17 |
-
Title = {{SciTail}: A Textual Entailment Dataset from Science Question Answering},
|
18 |
-
Year = {2018}
|
19 |
-
}
|
20 |
-
"""
|
21 |
-
|
22 |
-
# TODO(sciTail):
|
23 |
-
_DESCRIPTION = """\
|
24 |
-
The SciTail dataset is an entailment dataset created from multiple-choice science exams and web sentences. Each question
|
25 |
-
and the correct answer choice are converted into an assertive statement to form the hypothesis. We use information
|
26 |
-
retrieval to obtain relevant text from a large text corpus of web sentences, and use these sentences as a premise P. We
|
27 |
-
crowdsource the annotation of such premise-hypothesis pair as supports (entails) or not (neutral), in order to create
|
28 |
-
the SciTail dataset. The dataset contains 27,026 examples with 10,101 examples with entails label and 16,925 examples
|
29 |
-
with neutral label
|
30 |
-
"""
|
31 |
-
|
32 |
-
_URL = "http://data.allenai.org.s3.amazonaws.com/downloads/SciTailV1.1.zip"
|
33 |
-
|
34 |
-
|
35 |
-
class ScitailConfig(datasets.BuilderConfig):
|
36 |
-
|
37 |
-
"""BuilderConfig for Xquad"""
|
38 |
-
|
39 |
-
def __init__(self, **kwargs):
|
40 |
-
"""
|
41 |
-
|
42 |
-
Args:
|
43 |
-
**kwargs: keyword arguments forwarded to super.
|
44 |
-
"""
|
45 |
-
super(ScitailConfig, self).__init__(version=datasets.Version("1.1.0", ""), **kwargs)
|
46 |
-
|
47 |
-
|
48 |
-
class Scitail(datasets.GeneratorBasedBuilder):
|
49 |
-
"""TODO(sciTail): Short description of my dataset."""
|
50 |
-
|
51 |
-
# TODO(sciTail): Set up version.
|
52 |
-
VERSION = datasets.Version("1.1.0")
|
53 |
-
BUILDER_CONFIGS = [
|
54 |
-
ScitailConfig(
|
55 |
-
name="snli_format",
|
56 |
-
description="JSONL format used by SNLI with a JSON object corresponding to each entailment example in each line.",
|
57 |
-
),
|
58 |
-
ScitailConfig(
|
59 |
-
name="tsv_format", description="Tab-separated format with three columns: premise hypothesis label"
|
60 |
-
),
|
61 |
-
ScitailConfig(
|
62 |
-
name="dgem_format",
|
63 |
-
description="Tab-separated format used by the DGEM model: premise hypothesis label hypothesis graph structure",
|
64 |
-
),
|
65 |
-
ScitailConfig(
|
66 |
-
name="predictor_format",
|
67 |
-
description=textwrap.dedent(
|
68 |
-
"""\
|
69 |
-
AllenNLP predictors work only with JSONL format. This folder contains the SciTail train/dev/test in JSONL format
|
70 |
-
so that it can be loaded into the predictors. Each line is a JSON object with the following keys:
|
71 |
-
gold_label : the example label from {entails, neutral}
|
72 |
-
sentence1: the premise
|
73 |
-
sentence2: the hypothesis
|
74 |
-
sentence2_structure: structure from the hypothesis """
|
75 |
-
),
|
76 |
-
),
|
77 |
-
]
|
78 |
-
|
79 |
-
def _info(self):
|
80 |
-
# TODO(sciTail): Specifies the datasets.DatasetInfo object
|
81 |
-
if self.config.name == "snli_format":
|
82 |
-
return datasets.DatasetInfo(
|
83 |
-
# This is the description that will appear on the datasets page.
|
84 |
-
description=_DESCRIPTION,
|
85 |
-
# datasets.features.FeatureConnectors
|
86 |
-
features=datasets.Features(
|
87 |
-
{
|
88 |
-
"sentence1_binary_parse": datasets.Value("string"),
|
89 |
-
"sentence1_parse": datasets.Value("string"),
|
90 |
-
"sentence1": datasets.Value("string"),
|
91 |
-
"sentence2_parse": datasets.Value("string"),
|
92 |
-
"sentence2": datasets.Value("string"),
|
93 |
-
"annotator_labels": datasets.features.Sequence(datasets.Value("string")),
|
94 |
-
"gold_label": datasets.Value("string")
|
95 |
-
# These are the features of your dataset like images, labels ...
|
96 |
-
}
|
97 |
-
),
|
98 |
-
# If there's a common (input, target) tuple from the features,
|
99 |
-
# specify them here. They'll be used if as_supervised=True in
|
100 |
-
# builder.as_dataset.
|
101 |
-
supervised_keys=None,
|
102 |
-
# Homepage of the dataset for documentation
|
103 |
-
homepage="https://allenai.org/data/scitail",
|
104 |
-
citation=_CITATION,
|
105 |
-
)
|
106 |
-
elif self.config.name == "tsv_format":
|
107 |
-
return datasets.DatasetInfo(
|
108 |
-
# This is the description that will appear on the datasets page.
|
109 |
-
description=_DESCRIPTION,
|
110 |
-
# datasets.features.FeatureConnectors
|
111 |
-
features=datasets.Features(
|
112 |
-
{
|
113 |
-
"premise": datasets.Value("string"),
|
114 |
-
"hypothesis": datasets.Value("string"),
|
115 |
-
"label": datasets.Value("string")
|
116 |
-
# These are the features of your dataset like images, labels ...
|
117 |
-
}
|
118 |
-
),
|
119 |
-
# If there's a common (input, target) tuple from the features,
|
120 |
-
# specify them here. They'll be used if as_supervised=True in
|
121 |
-
# builder.as_dataset.
|
122 |
-
supervised_keys=None,
|
123 |
-
# Homepage of the dataset for documentation
|
124 |
-
homepage="https://allenai.org/data/scitail",
|
125 |
-
citation=_CITATION,
|
126 |
-
)
|
127 |
-
elif self.config.name == "predictor_format":
|
128 |
-
return datasets.DatasetInfo(
|
129 |
-
# This is the description that will appear on the datasets page.
|
130 |
-
description=_DESCRIPTION,
|
131 |
-
# datasets.features.FeatureConnectors
|
132 |
-
features=datasets.Features(
|
133 |
-
{
|
134 |
-
"answer": datasets.Value("string"),
|
135 |
-
"sentence2_structure": datasets.Value("string"),
|
136 |
-
"sentence1": datasets.Value("string"),
|
137 |
-
"sentence2": datasets.Value("string"),
|
138 |
-
"gold_label": datasets.Value("string"),
|
139 |
-
"question": datasets.Value("string")
|
140 |
-
# These are the features of your dataset like images, labels ...
|
141 |
-
}
|
142 |
-
),
|
143 |
-
# If there's a common (input, target) tuple from the features,
|
144 |
-
# specify them here. They'll be used if as_supervised=True in
|
145 |
-
# builder.as_dataset.
|
146 |
-
supervised_keys=None,
|
147 |
-
# Homepage of the dataset for documentation
|
148 |
-
homepage="https://allenai.org/data/scitail",
|
149 |
-
citation=_CITATION,
|
150 |
-
)
|
151 |
-
elif self.config.name == "dgem_format":
|
152 |
-
return datasets.DatasetInfo(
|
153 |
-
# This is the description that will appear on the datasets page.
|
154 |
-
description=_DESCRIPTION,
|
155 |
-
# datasets.features.FeatureConnectors
|
156 |
-
features=datasets.Features(
|
157 |
-
{
|
158 |
-
"premise": datasets.Value("string"),
|
159 |
-
"hypothesis": datasets.Value("string"),
|
160 |
-
"label": datasets.Value("string"),
|
161 |
-
"hypothesis_graph_structure": datasets.Value("string")
|
162 |
-
# These are the features of your dataset like images, labels ...
|
163 |
-
}
|
164 |
-
),
|
165 |
-
# If there's a common (input, target) tuple from the features,
|
166 |
-
# specify them here. They'll be used if as_supervised=True in
|
167 |
-
# builder.as_dataset.
|
168 |
-
supervised_keys=None,
|
169 |
-
# Homepage of the dataset for documentation
|
170 |
-
homepage="https://allenai.org/data/scitail",
|
171 |
-
citation=_CITATION,
|
172 |
-
)
|
173 |
-
|
174 |
-
def _split_generators(self, dl_manager):
|
175 |
-
"""Returns SplitGenerators."""
|
176 |
-
# TODO(sciTail): Downloads the data and defines the splits
|
177 |
-
# dl_manager is a datasets.download.DownloadManager that can be used to
|
178 |
-
# download and extract URLs
|
179 |
-
dl_dir = dl_manager.download_and_extract(_URL)
|
180 |
-
data_dir = os.path.join(dl_dir, "SciTailV1.1")
|
181 |
-
snli = os.path.join(data_dir, "snli_format")
|
182 |
-
dgem = os.path.join(data_dir, "dgem_format")
|
183 |
-
tsv = os.path.join(data_dir, "tsv_format")
|
184 |
-
predictor = os.path.join(data_dir, "predictor_format")
|
185 |
-
if self.config.name == "snli_format":
|
186 |
-
return [
|
187 |
-
datasets.SplitGenerator(
|
188 |
-
name=datasets.Split.TRAIN,
|
189 |
-
# These kwargs will be passed to _generate_examples
|
190 |
-
gen_kwargs={"filepath": os.path.join(snli, "scitail_1.0_train.txt")},
|
191 |
-
),
|
192 |
-
datasets.SplitGenerator(
|
193 |
-
name=datasets.Split.TEST,
|
194 |
-
# These kwargs will be passed to _generate_examples
|
195 |
-
gen_kwargs={"filepath": os.path.join(snli, "scitail_1.0_test.txt")},
|
196 |
-
),
|
197 |
-
datasets.SplitGenerator(
|
198 |
-
name=datasets.Split.VALIDATION,
|
199 |
-
# These kwargs will be passed to _generate_examples
|
200 |
-
gen_kwargs={"filepath": os.path.join(snli, "scitail_1.0_dev.txt")},
|
201 |
-
),
|
202 |
-
]
|
203 |
-
elif self.config.name == "tsv_format":
|
204 |
-
return [
|
205 |
-
datasets.SplitGenerator(
|
206 |
-
name=datasets.Split.TRAIN,
|
207 |
-
# These kwargs will be passed to _generate_examples
|
208 |
-
gen_kwargs={"filepath": os.path.join(tsv, "scitail_1.0_train.tsv")},
|
209 |
-
),
|
210 |
-
datasets.SplitGenerator(
|
211 |
-
name=datasets.Split.TEST,
|
212 |
-
# These kwargs will be passed to _generate_examples
|
213 |
-
gen_kwargs={"filepath": os.path.join(tsv, "scitail_1.0_test.tsv")},
|
214 |
-
),
|
215 |
-
datasets.SplitGenerator(
|
216 |
-
name=datasets.Split.VALIDATION,
|
217 |
-
# These kwargs will be passed to _generate_examples
|
218 |
-
gen_kwargs={"filepath": os.path.join(tsv, "scitail_1.0_dev.tsv")},
|
219 |
-
),
|
220 |
-
]
|
221 |
-
elif self.config.name == "predictor_format":
|
222 |
-
return [
|
223 |
-
datasets.SplitGenerator(
|
224 |
-
name=datasets.Split.TRAIN,
|
225 |
-
# These kwargs will be passed to _generate_examples
|
226 |
-
gen_kwargs={"filepath": os.path.join(predictor, "scitail_1.0_structure_train.jsonl")},
|
227 |
-
),
|
228 |
-
datasets.SplitGenerator(
|
229 |
-
name=datasets.Split.TEST,
|
230 |
-
# These kwargs will be passed to _generate_examples
|
231 |
-
gen_kwargs={"filepath": os.path.join(predictor, "scitail_1.0_structure_test.jsonl")},
|
232 |
-
),
|
233 |
-
datasets.SplitGenerator(
|
234 |
-
name=datasets.Split.VALIDATION,
|
235 |
-
# These kwargs will be passed to _generate_examples
|
236 |
-
gen_kwargs={"filepath": os.path.join(predictor, "scitail_1.0_structure_dev.jsonl")},
|
237 |
-
),
|
238 |
-
]
|
239 |
-
elif self.config.name == "dgem_format":
|
240 |
-
return [
|
241 |
-
datasets.SplitGenerator(
|
242 |
-
name=datasets.Split.TRAIN,
|
243 |
-
# These kwargs will be passed to _generate_examples
|
244 |
-
gen_kwargs={"filepath": os.path.join(dgem, "scitail_1.0_structure_train.tsv")},
|
245 |
-
),
|
246 |
-
datasets.SplitGenerator(
|
247 |
-
name=datasets.Split.TEST,
|
248 |
-
# These kwargs will be passed to _generate_examples
|
249 |
-
gen_kwargs={"filepath": os.path.join(dgem, "scitail_1.0_structure_test.tsv")},
|
250 |
-
),
|
251 |
-
datasets.SplitGenerator(
|
252 |
-
name=datasets.Split.VALIDATION,
|
253 |
-
# These kwargs will be passed to _generate_examples
|
254 |
-
gen_kwargs={"filepath": os.path.join(dgem, "scitail_1.0_structure_dev.tsv")},
|
255 |
-
),
|
256 |
-
]
|
257 |
-
|
258 |
-
def _generate_examples(self, filepath):
|
259 |
-
"""Yields examples."""
|
260 |
-
# TODO(sciTail): Yields (key, example) tuples from the dataset
|
261 |
-
with open(filepath, encoding="utf-8") as f:
|
262 |
-
if self.config.name == "snli_format":
|
263 |
-
for id_, row in enumerate(f):
|
264 |
-
data = json.loads(row)
|
265 |
-
|
266 |
-
yield id_, {
|
267 |
-
"sentence1_binary_parse": data["sentence1_binary_parse"],
|
268 |
-
"sentence1_parse": data["sentence1_parse"],
|
269 |
-
"sentence1": data["sentence1"],
|
270 |
-
"sentence2_parse": data["sentence2_parse"],
|
271 |
-
"sentence2": data["sentence2"],
|
272 |
-
"annotator_labels": data["annotator_labels"],
|
273 |
-
"gold_label": data["gold_label"],
|
274 |
-
}
|
275 |
-
elif self.config.name == "tsv_format":
|
276 |
-
data = csv.reader(f, delimiter="\t")
|
277 |
-
for id_, row in enumerate(data):
|
278 |
-
yield id_, {"premise": row[0], "hypothesis": row[1], "label": row[2]}
|
279 |
-
elif self.config.name == "dgem_format":
|
280 |
-
data = csv.reader(f, delimiter="\t")
|
281 |
-
for id_, row in enumerate(data):
|
282 |
-
yield id_, {
|
283 |
-
"premise": row[0],
|
284 |
-
"hypothesis": row[1],
|
285 |
-
"label": row[2],
|
286 |
-
"hypothesis_graph_structure": row[3],
|
287 |
-
}
|
288 |
-
elif self.config.name == "predictor_format":
|
289 |
-
for id_, row in enumerate(f):
|
290 |
-
data = json.loads(row)
|
291 |
-
yield id_, {
|
292 |
-
"answer": data["answer"],
|
293 |
-
"sentence2_structure": data["sentence2_structure"],
|
294 |
-
"sentence1": data["sentence1"],
|
295 |
-
"sentence2": data["sentence2"],
|
296 |
-
"gold_label": data["gold_label"],
|
297 |
-
"question": data["question"],
|
298 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|