gabrielaltay
commited on
Commit
•
a5b3217
1
Parent(s):
509b49e
upload hubscripts/evidence_inference_hub.py to hub from bigbio repo
Browse files- evidence_inference.py +293 -0
evidence_inference.py
ADDED
@@ -0,0 +1,293 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 |
+
The dataset consists of biomedical articles describing randomized control trials (RCTs)
|
18 |
+
that compare multiple treatments. Each of these articles will have multiple questions,
|
19 |
+
or 'prompts' associated with them. These prompts will ask about the relationship between
|
20 |
+
an intervention and comparator with respect to an outcome, as reported in the trial.
|
21 |
+
For example, a prompt may ask about the reported effects of aspirin as compared to placebo
|
22 |
+
on the duration of headaches.
|
23 |
+
For the sake of this task, we assume that a particular article will report that the intervention of interest either
|
24 |
+
significantly increased, significantly decreased or had significant effect on the outcome, relative to the comparator.
|
25 |
+
"""
|
26 |
+
|
27 |
+
import os
|
28 |
+
from typing import Dict, List, Tuple
|
29 |
+
|
30 |
+
import datasets
|
31 |
+
import pandas as pd
|
32 |
+
|
33 |
+
from .bigbiohub import qa_features
|
34 |
+
from .bigbiohub import BigBioConfig
|
35 |
+
from .bigbiohub import Tasks
|
36 |
+
|
37 |
+
_LANGUAGES = ['English']
|
38 |
+
_PUBMED = True
|
39 |
+
_LOCAL = False
|
40 |
+
_CITATION = """\
|
41 |
+
@inproceedings{deyoung-etal-2020-evidence,
|
42 |
+
title = "Evidence Inference 2.0: More Data, Better Models",
|
43 |
+
author = "DeYoung, Jay and
|
44 |
+
Lehman, Eric and
|
45 |
+
Nye, Benjamin and
|
46 |
+
Marshall, Iain and
|
47 |
+
Wallace, Byron C.",
|
48 |
+
booktitle = "Proceedings of the 19th SIGBioMed Workshop on Biomedical Language Processing",
|
49 |
+
month = jul,
|
50 |
+
year = "2020",
|
51 |
+
address = "Online",
|
52 |
+
publisher = "Association for Computational Linguistics",
|
53 |
+
url = "https://www.aclweb.org/anthology/2020.bionlp-1.13",
|
54 |
+
pages = "123--132",
|
55 |
+
}
|
56 |
+
"""
|
57 |
+
|
58 |
+
_DATASETNAME = "evidence_inference"
|
59 |
+
_DISPLAYNAME = "Evidence Inference 2.0"
|
60 |
+
|
61 |
+
_DESCRIPTION = """\
|
62 |
+
The dataset consists of biomedical articles describing randomized control trials (RCTs) that compare multiple
|
63 |
+
treatments. Each of these articles will have multiple questions, or 'prompts' associated with them.
|
64 |
+
These prompts will ask about the relationship between an intervention and comparator with respect to an outcome,
|
65 |
+
as reported in the trial. For example, a prompt may ask about the reported effects of aspirin as compared
|
66 |
+
to placebo on the duration of headaches. For the sake of this task, we assume that a particular article
|
67 |
+
will report that the intervention of interest either significantly increased, significantly decreased
|
68 |
+
or had significant effect on the outcome, relative to the comparator.
|
69 |
+
"""
|
70 |
+
|
71 |
+
_HOMEPAGE = "https://github.com/jayded/evidence-inference"
|
72 |
+
|
73 |
+
_LICENSE = 'MIT License'
|
74 |
+
|
75 |
+
_URLS = {
|
76 |
+
_DATASETNAME: "http://evidence-inference.ebm-nlp.com/v2.0.tar.gz",
|
77 |
+
}
|
78 |
+
|
79 |
+
_SUPPORTED_TASKS = [Tasks.QUESTION_ANSWERING]
|
80 |
+
|
81 |
+
_SOURCE_VERSION = "2.0.0"
|
82 |
+
|
83 |
+
_BIGBIO_VERSION = "1.0.0"
|
84 |
+
|
85 |
+
QA_CHOICES = [
|
86 |
+
"significantly increased",
|
87 |
+
"no significant difference",
|
88 |
+
"significantly decreased",
|
89 |
+
]
|
90 |
+
|
91 |
+
# Some examples are removed due to comments on the dataset's github page
|
92 |
+
# https://github.com/jayded/evidence-inference/blob/master/annotations/README.md#caveat
|
93 |
+
|
94 |
+
INCORRECT_PROMPT_IDS = set([
|
95 |
+
911, 912, 1262, 1261, 3044, 3248, 3111, 3620, 4308, 4490, 4491, 4324,
|
96 |
+
4325, 4492, 4824, 5000, 5001, 5002, 5046, 5047, 4948, 5639, 5710, 5752,
|
97 |
+
5775, 5782, 5841, 5843, 5861, 5862, 5863, 5964, 5965, 5966, 5975, 4807,
|
98 |
+
5776, 5777, 5778, 5779, 5780, 5781, 6034, 6065, 6066, 6666, 6667, 6668,
|
99 |
+
6669, 7040, 7042, 7944, 8590, 8605, 8606, 8639, 8640, 8745, 8747, 8749,
|
100 |
+
8877, 8878, 8593, 8631, 8635, 8884, 8886, 8773, 10032, 10035, 8876, 8875,
|
101 |
+
8885, 8917, 8921, 8118, 10885, 10886, 10887, 10888, 10889, 10890
|
102 |
+
])
|
103 |
+
|
104 |
+
QUESTIONABLE_PROMPT_IDS = set([
|
105 |
+
7811, 7812, 7813, 7814, 7815, 8197, 8198, 8199,
|
106 |
+
8200, 8201, 9429, 9430, 9431, 8536, 9432
|
107 |
+
])
|
108 |
+
|
109 |
+
SOMEWHAT_MALFORMED_PROMPT_IDS = set([
|
110 |
+
3514, 346, 5037, 4715, 8767, 9295, 9297, 8870, 9862
|
111 |
+
])
|
112 |
+
|
113 |
+
SKIP_PROMPT_IDS = INCORRECT_PROMPT_IDS | QUESTIONABLE_PROMPT_IDS | SOMEWHAT_MALFORMED_PROMPT_IDS
|
114 |
+
|
115 |
+
|
116 |
+
class EvidenceInferenceDataset(datasets.GeneratorBasedBuilder):
|
117 |
+
f"""{_DESCRIPTION}"""
|
118 |
+
|
119 |
+
SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
|
120 |
+
BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION)
|
121 |
+
|
122 |
+
BUILDER_CONFIGS = [
|
123 |
+
BigBioConfig(
|
124 |
+
name="evidence-inference_source",
|
125 |
+
version=SOURCE_VERSION,
|
126 |
+
description="evidence-inference source schema",
|
127 |
+
schema="source",
|
128 |
+
subset_id="evidence-inference",
|
129 |
+
),
|
130 |
+
BigBioConfig(
|
131 |
+
name="evidence-inference_bigbio_qa",
|
132 |
+
version=BIGBIO_VERSION,
|
133 |
+
description="evidence-inference BigBio schema",
|
134 |
+
schema="bigbio_qa",
|
135 |
+
subset_id="evidence-inference",
|
136 |
+
),
|
137 |
+
]
|
138 |
+
|
139 |
+
DEFAULT_CONFIG_NAME = "evidence-inference_source"
|
140 |
+
|
141 |
+
def _info(self) -> datasets.DatasetInfo:
|
142 |
+
if self.config.schema == "source":
|
143 |
+
features = datasets.Features(
|
144 |
+
{
|
145 |
+
"id": datasets.Value("int64"),
|
146 |
+
"prompt_id": datasets.Value("int64"),
|
147 |
+
"pmcid": datasets.Value("int64"),
|
148 |
+
"label": datasets.Value("string"),
|
149 |
+
"evidence": datasets.Value("string"),
|
150 |
+
"intervention": datasets.Value("string"),
|
151 |
+
"comparator": datasets.Value("string"),
|
152 |
+
"outcome": datasets.Value("string"),
|
153 |
+
}
|
154 |
+
)
|
155 |
+
|
156 |
+
elif self.config.schema == "bigbio_qa":
|
157 |
+
features = qa_features
|
158 |
+
|
159 |
+
return datasets.DatasetInfo(
|
160 |
+
description=_DESCRIPTION,
|
161 |
+
features=features,
|
162 |
+
homepage=_HOMEPAGE,
|
163 |
+
license=str(_LICENSE),
|
164 |
+
citation=_CITATION,
|
165 |
+
)
|
166 |
+
|
167 |
+
def _split_generators(self, dl_manager) -> List[datasets.SplitGenerator]:
|
168 |
+
"""Returns SplitGenerators."""
|
169 |
+
|
170 |
+
urls = _URLS[_DATASETNAME]
|
171 |
+
data_dir = dl_manager.download_and_extract(urls)
|
172 |
+
|
173 |
+
return [
|
174 |
+
datasets.SplitGenerator(
|
175 |
+
name=datasets.Split.TRAIN,
|
176 |
+
gen_kwargs={
|
177 |
+
"filepaths": [
|
178 |
+
os.path.join(data_dir, "annotations_merged.csv"),
|
179 |
+
os.path.join(data_dir, "prompts_merged.csv"),
|
180 |
+
],
|
181 |
+
"datapath": os.path.join(data_dir, "txt_files"),
|
182 |
+
"split": "train",
|
183 |
+
"datadir": data_dir,
|
184 |
+
},
|
185 |
+
),
|
186 |
+
datasets.SplitGenerator(
|
187 |
+
name=datasets.Split.VALIDATION,
|
188 |
+
gen_kwargs={
|
189 |
+
"filepaths": [
|
190 |
+
os.path.join(data_dir, "annotations_merged.csv"),
|
191 |
+
os.path.join(data_dir, "prompts_merged.csv"),
|
192 |
+
],
|
193 |
+
"datapath": os.path.join(data_dir, "txt_files"),
|
194 |
+
"split": "validation",
|
195 |
+
"datadir": data_dir,
|
196 |
+
},
|
197 |
+
),
|
198 |
+
datasets.SplitGenerator(
|
199 |
+
name=datasets.Split.TEST,
|
200 |
+
gen_kwargs={
|
201 |
+
"filepaths": [
|
202 |
+
os.path.join(data_dir, "annotations_merged.csv"),
|
203 |
+
os.path.join(data_dir, "prompts_merged.csv"),
|
204 |
+
],
|
205 |
+
"datapath": os.path.join(data_dir, "txt_files"),
|
206 |
+
"split": "test",
|
207 |
+
"datadir": data_dir,
|
208 |
+
},
|
209 |
+
),
|
210 |
+
]
|
211 |
+
|
212 |
+
def _generate_examples(
|
213 |
+
self, filepaths, datapath, split, datadir
|
214 |
+
) -> Tuple[int, Dict]:
|
215 |
+
"""Yields examples as (key, example) tuples."""
|
216 |
+
with open(f"{datadir}/splits/{split}_article_ids.txt", "r") as f:
|
217 |
+
ids = [int(i.strip()) for i in f.readlines()]
|
218 |
+
prompts = pd.read_csv(filepaths[-1], encoding="utf8")
|
219 |
+
prompts = prompts[prompts["PMCID"].isin(ids)]
|
220 |
+
|
221 |
+
annotations = pd.read_csv(filepaths[0], encoding="utf8").set_index("PromptID")
|
222 |
+
evidences = pd.read_csv(filepaths[0], encoding="utf8").set_index("PMCID")
|
223 |
+
evidences = evidences[evidences["Evidence Start"] != -1]
|
224 |
+
uid = 0
|
225 |
+
|
226 |
+
def lookup(df: pd.DataFrame, id, col) -> str:
|
227 |
+
try:
|
228 |
+
label = df.loc[id][col]
|
229 |
+
if isinstance(label, pd.Series):
|
230 |
+
return label.values[0]
|
231 |
+
else:
|
232 |
+
return label
|
233 |
+
except KeyError:
|
234 |
+
return -1
|
235 |
+
|
236 |
+
def extract_evidence(doc_id, start, end):
|
237 |
+
p = f"{datapath}/PMC{doc_id}.txt"
|
238 |
+
with open(p, "r") as f:
|
239 |
+
return f.read()[start:end]
|
240 |
+
|
241 |
+
|
242 |
+
for key, sample in prompts.iterrows():
|
243 |
+
|
244 |
+
pid = sample["PromptID"]
|
245 |
+
pmcid = sample["PMCID"]
|
246 |
+
label = lookup(annotations, pid, "Label")
|
247 |
+
start = lookup(evidences, pmcid, "Evidence Start")
|
248 |
+
end = lookup(evidences, pmcid, "Evidence End")
|
249 |
+
|
250 |
+
if pid in SKIP_PROMPT_IDS:
|
251 |
+
continue
|
252 |
+
|
253 |
+
if label == -1:
|
254 |
+
continue
|
255 |
+
|
256 |
+
evidence = extract_evidence(pmcid, start, end)
|
257 |
+
|
258 |
+
if self.config.schema == "source":
|
259 |
+
|
260 |
+
feature_dict = {
|
261 |
+
"id": uid,
|
262 |
+
"pmcid": pmcid,
|
263 |
+
"prompt_id": pid,
|
264 |
+
"intervention": sample["Intervention"],
|
265 |
+
"comparator": sample["Comparator"],
|
266 |
+
"outcome": sample["Outcome"],
|
267 |
+
"evidence": evidence,
|
268 |
+
"label": label,
|
269 |
+
}
|
270 |
+
|
271 |
+
uid += 1
|
272 |
+
yield key, feature_dict
|
273 |
+
|
274 |
+
elif self.config.schema == "bigbio_qa":
|
275 |
+
|
276 |
+
context = evidence
|
277 |
+
question = (
|
278 |
+
f"Compared to {sample['Comparator']} "
|
279 |
+
f"what was the result of {sample['Intervention']} on {sample['Outcome']}?"
|
280 |
+
)
|
281 |
+
feature_dict = {
|
282 |
+
"id": uid,
|
283 |
+
"question_id": pid,
|
284 |
+
"document_id": pmcid,
|
285 |
+
"question": question,
|
286 |
+
"type": "multiple_choice",
|
287 |
+
"choices": QA_CHOICES,
|
288 |
+
"context": context,
|
289 |
+
"answer": [label],
|
290 |
+
}
|
291 |
+
|
292 |
+
uid += 1
|
293 |
+
yield key, feature_dict
|