Datasets:

Languages:
English
License:
gabrielaltay commited on
Commit
4a683e0
·
1 Parent(s): f6b8ff4

upload hubscripts/mediqa_nli_hub.py to hub from bigbio repo

Browse files
Files changed (1) hide show
  1. mediqa_nli.py +200 -0
mediqa_nli.py ADDED
@@ -0,0 +1,200 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
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
+ """
17
+ Natural Language Inference (NLI) is the task of determining whether a given hypothesis can be
18
+ inferred from a given premise. Also known as Recognizing Textual Entailment (RTE), this task has
19
+ enjoyed popularity among researchers for some time. However, almost all datasets for this task
20
+ focused on open domain data such as as news texts, blogs, and so on. To address this gap, the MedNLI
21
+ dataset was created for language inference in the medical domain. MedNLI is a derived dataset with
22
+ data sourced from MIMIC-III v1.4. In order to stimulate research for this problem, a shared task on
23
+ Medical Inference and Question Answering (MEDIQA) was organized at the workshop for biomedical
24
+ natural language processing (BioNLP) 2019. The dataset provided herein is a test set of 405 premise
25
+ hypothesis pairs for the NLI challenge in the MEDIQA shared task. Participants of the shared task
26
+ are expected to use the MedNLI data for development of their models and this dataset was used as an
27
+ unseen dataset for scoring each participant submission.
28
+
29
+ The files comprising this dataset must be on the users local machine in a single directory that is
30
+ passed to `datasets.load_datset` via the `data_dir` kwarg. This loader script will read the archive
31
+ files directly (i.e. the user should not uncompress, untar or unzip any of the files). For example,
32
+ if `data_dir` is `"mediqa_nli"` it should contain the following files:
33
+
34
+ mediqa_nli
35
+ ├── mednli-for-shared-task-at-acl-bionlp-2019-1.0.1.zip
36
+ """
37
+
38
+ import json
39
+ import os
40
+ from typing import Dict, List, Tuple
41
+
42
+ import datasets
43
+ import pandas as pd
44
+
45
+ from .bigbiohub import entailment_features
46
+ from .bigbiohub import BigBioConfig
47
+ from .bigbiohub import Tasks
48
+
49
+ _LANGUAGES = ['English']
50
+ _PUBMED = False
51
+ _LOCAL = True
52
+ _CITATION = """\
53
+ @misc{https://doi.org/10.13026/gtv4-g455,
54
+ title = {MedNLI for Shared Task at ACL BioNLP 2019},
55
+ author = {Shivade, Chaitanya},
56
+ year = 2019,
57
+ publisher = {physionet.org},
58
+ doi = {10.13026/GTV4-G455},
59
+ url = {https://physionet.org/content/mednli-bionlp19/}
60
+ }
61
+
62
+ """
63
+
64
+ _DATASETNAME = "mediqa_nli"
65
+ _DISPLAYNAME = "MEDIQA NLI"
66
+
67
+ _DESCRIPTION = """\
68
+ Natural Language Inference (NLI) is the task of determining whether a given hypothesis can be
69
+ inferred from a given premise. Also known as Recognizing Textual Entailment (RTE), this task has
70
+ enjoyed popularity among researchers for some time. However, almost all datasets for this task
71
+ focused on open domain data such as as news texts, blogs, and so on. To address this gap, the MedNLI
72
+ dataset was created for language inference in the medical domain. MedNLI is a derived dataset with
73
+ data sourced from MIMIC-III v1.4. In order to stimulate research for this problem, a shared task on
74
+ Medical Inference and Question Answering (MEDIQA) was organized at the workshop for biomedical
75
+ natural language processing (BioNLP) 2019. The dataset provided herein is a test set of 405 premise
76
+ hypothesis pairs for the NLI challenge in the MEDIQA shared task. Participants of the shared task
77
+ are expected to use the MedNLI data for development of their models and this dataset was used as an
78
+ unseen dataset for scoring each participant submission.
79
+ """
80
+
81
+
82
+ _HOMEPAGE = "https://physionet.org/content/mednli-bionlp19/1.0.1/"
83
+
84
+ _LICENSE = 'PhysioNet Credentialed Health Data License'
85
+
86
+ _URLS = {}
87
+
88
+ _SUPPORTED_TASKS = [Tasks.TEXTUAL_ENTAILMENT]
89
+
90
+ _SOURCE_VERSION = "1.0.1"
91
+ _BIGBIO_VERSION = "1.0.0"
92
+
93
+
94
+ class MEDIQANLIDataset(datasets.GeneratorBasedBuilder):
95
+ """MEDIQA NLI"""
96
+
97
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
98
+ BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION)
99
+
100
+ BUILDER_CONFIGS = [
101
+ BigBioConfig(
102
+ name="mediqa_nli_source",
103
+ version=SOURCE_VERSION,
104
+ description="MEDIQA NLI source schema",
105
+ schema="source",
106
+ subset_id="mediqa_nli",
107
+ ),
108
+ BigBioConfig(
109
+ name="mediqa_nli_bigbio_te",
110
+ version=BIGBIO_VERSION,
111
+ description="MEDIQA NLI BigBio schema",
112
+ schema="bigbio_te",
113
+ subset_id="mediqa_nli",
114
+ ),
115
+ ]
116
+
117
+ DEFAULT_CONFIG_NAME = "mediqa_nli_source"
118
+
119
+ def _info(self) -> datasets.DatasetInfo:
120
+
121
+ if self.config.schema == "source":
122
+ features = datasets.Features(
123
+ {
124
+ "pairID": datasets.Value("string"),
125
+ "gold_label": datasets.Value("string"),
126
+ "sentence1": datasets.Value("string"),
127
+ "sentence2": datasets.Value("string"),
128
+ "sentence1_parse": datasets.Value("string"),
129
+ "sentence2_parse": datasets.Value("string"),
130
+ "sentence1_binary_parse": datasets.Value("string"),
131
+ "sentence2_binary_parse": datasets.Value("string"),
132
+ }
133
+ )
134
+
135
+ elif self.config.schema == "bigbio_te":
136
+ features = entailment_features
137
+
138
+ return datasets.DatasetInfo(
139
+ description=_DESCRIPTION,
140
+ features=features,
141
+ homepage=_HOMEPAGE,
142
+ license=str(_LICENSE),
143
+ citation=_CITATION,
144
+ )
145
+
146
+ def _split_generators(self, dl_manager) -> List[datasets.SplitGenerator]:
147
+ if self.config.data_dir is None:
148
+ raise ValueError(
149
+ "This is a local dataset. Please pass the data_dir kwarg to load_dataset."
150
+ )
151
+ else:
152
+ extract_dir = dl_manager.extract(
153
+ os.path.join(
154
+ self.config.data_dir,
155
+ "mednli-for-shared-task-at-acl-bionlp-2019-1.0.1.zip",
156
+ )
157
+ )
158
+ data_dir = os.path.join(
159
+ extract_dir, "mednli-for-shared-task-at-acl-bionlp-2019-1.0.1"
160
+ )
161
+
162
+ return [
163
+ datasets.SplitGenerator(
164
+ name=datasets.Split.TEST,
165
+ gen_kwargs={
166
+ "examples_filepath": os.path.join(
167
+ data_dir, "mednli_bionlp19_shared_task.jsonl"
168
+ ),
169
+ "ground_truth_filepath": os.path.join(
170
+ data_dir, "mednli_bionlp19_shared_task_ground_truth.csv"
171
+ ),
172
+ "split": "test",
173
+ },
174
+ ),
175
+ ]
176
+
177
+ def _generate_examples(
178
+ self, examples_filepath: str, ground_truth_filepath: str, split: str
179
+ ) -> Tuple[int, Dict]:
180
+
181
+ ground_truth = pd.read_csv(
182
+ ground_truth_filepath, index_col=0, squeeze=True
183
+ ).to_dict()
184
+ with open(examples_filepath, "r") as f:
185
+ if self.config.schema == "source":
186
+ for line in f:
187
+ json_line = json.loads(line)
188
+ json_line["gold_label"] = ground_truth[json_line["pairID"]]
189
+ yield json_line["pairID"], json_line
190
+
191
+ elif self.config.schema == "bigbio_te":
192
+ for line in f:
193
+ json_line = json.loads(line)
194
+ entailment_example = {
195
+ "id": json_line["pairID"],
196
+ "premise": json_line["sentence1"],
197
+ "hypothesis": json_line["sentence2"],
198
+ "label": ground_truth[json_line["pairID"]],
199
+ }
200
+ yield json_line["pairID"], entailment_example