Convert dataset to Parquet (#1)
Browse files- Convert dataset to Parquet (c06146153a29c5c8d3a4cd27878ea7a2919193a2)
- Add 'strains' config data files (1db987caa6bc9a8806375afd5ec5fbe27311bd4d)
- Delete loading script (1ee40cc7d7e54eb60574afff4f9e0559ff802c48)
- Delete data file (725df46de952c2efd5b64e3c712dbd703dbb18aa)
- .gitignore +0 -2
- README.md +12 -2
- pathogens/train-00000-of-00001.parquet +3 -0
- strains/train-00000-of-00001.parquet +3 -0
- tea_curated.py +0 -238
.gitignore
DELETED
@@ -1,2 +0,0 @@
|
|
1 |
-
flake.nix
|
2 |
-
flake.lock
|
|
|
|
|
|
README.md
CHANGED
@@ -37,7 +37,7 @@ dataset_info:
|
|
37 |
- name: train
|
38 |
num_bytes: 50995544
|
39 |
num_examples: 505
|
40 |
-
download_size:
|
41 |
dataset_size: 50995544
|
42 |
- config_name: strains
|
43 |
features:
|
@@ -70,6 +70,16 @@ dataset_info:
|
|
70 |
- name: train
|
71 |
num_bytes: 41153821
|
72 |
num_examples: 400
|
73 |
-
download_size:
|
74 |
dataset_size: 41153821
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
---
|
|
|
37 |
- name: train
|
38 |
num_bytes: 50995544
|
39 |
num_examples: 505
|
40 |
+
download_size: 11318610
|
41 |
dataset_size: 50995544
|
42 |
- config_name: strains
|
43 |
features:
|
|
|
70 |
- name: train
|
71 |
num_bytes: 41153821
|
72 |
num_examples: 400
|
73 |
+
download_size: 8836267
|
74 |
dataset_size: 41153821
|
75 |
+
configs:
|
76 |
+
- config_name: pathogens
|
77 |
+
data_files:
|
78 |
+
- split: train
|
79 |
+
path: pathogens/train-*
|
80 |
+
default: true
|
81 |
+
- config_name: strains
|
82 |
+
data_files:
|
83 |
+
- split: train
|
84 |
+
path: strains/train-*
|
85 |
---
|
pathogens/train-00000-of-00001.parquet
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c3d6e32db9118f41876ac2144fa5fc951c9985da44eaef244ea5fdb81e8d7a88
|
3 |
+
size 11318610
|
strains/train-00000-of-00001.parquet
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:fd15182d90d212137f5c72604a1021969e7ac266d395312bd9db692ed02693c1
|
3 |
+
size 8836267
|
tea_curated.py
DELETED
@@ -1,238 +0,0 @@
|
|
1 |
-
# -*- coding: utf-8 -*-
|
2 |
-
"""
|
3 |
-
TEA: Taxonomic Entity Augmentation (curated raw corpus)
|
4 |
-
|
5 |
-
Usage
|
6 |
-
-----
|
7 |
-
>>> from datasets import load_dataset
|
8 |
-
>>> ds = load_dataset("tznurmin/tea_curated",
|
9 |
-
... "pathogens",
|
10 |
-
... trust_remote_code=True,
|
11 |
-
... split="train")
|
12 |
-
"""
|
13 |
-
|
14 |
-
import json
|
15 |
-
import logging
|
16 |
-
import re
|
17 |
-
from pathlib import Path
|
18 |
-
|
19 |
-
import datasets
|
20 |
-
|
21 |
-
logger = logging.getLogger(__name__)
|
22 |
-
|
23 |
-
# ---------------------------------------------------------------------------
|
24 |
-
# Hub metadata
|
25 |
-
# ---------------------------------------------------------------------------
|
26 |
-
_HOMEPAGE = "https://github.com/tznurmin/TEA_curated_data"
|
27 |
-
|
28 |
-
_LICENSE = (
|
29 |
-
"MIT (curation JSON annotations) and mixed Creative-Commons licences "
|
30 |
-
"(CC-BY / CC-BY-SA / CC-BY-NC / CC-BY-NC-SA) for article texts"
|
31 |
-
)
|
32 |
-
|
33 |
-
_CITATION = r"""
|
34 |
-
@misc{nurminen2025tea,
|
35 |
-
author = {Nurminen, Toni and Kim, John and Si, Tong and Chang, Matthew Wook},
|
36 |
-
title = {TEA: Taxonomic Entity Augmentation for Biological Literature},
|
37 |
-
year = {2025},
|
38 |
-
howpublished = {\url{https://github.com/tznurmin/TEA_curated_data}},
|
39 |
-
note = {Dataset available at \url{https://github.com/tznurmin/TEA_curated_data}}
|
40 |
-
}
|
41 |
-
"""
|
42 |
-
|
43 |
-
_DESCRIPTION = (
|
44 |
-
"Taxonomic Entity Augmentation (TEA) curated dataset. It contains the full "
|
45 |
-
"text of biomedical articles (under their original Creative-Commons licences) "
|
46 |
-
"and curation JSON files that mark pathogen / strain / species entities."
|
47 |
-
"Two configs are provided:\n\n"
|
48 |
-
"1: **pathogens** – commensals, probiotics, opportunistics, negatives, strains, pathogens\n"
|
49 |
-
"2: **strains** – strains, species, negatives\n\n"
|
50 |
-
"Each token is annotated in **IOB2** format (B-XXX, I-XXX, O). Per-record "
|
51 |
-
"licence information and attribution text are included."
|
52 |
-
)
|
53 |
-
|
54 |
-
_URL = "https://github.com/tznurmin/TEA_curated_data/archive/refs/tags/v1.1.tar.gz"
|
55 |
-
|
56 |
-
# ---------------------------------------------------------------------------
|
57 |
-
# IOB2 label sets
|
58 |
-
# ---------------------------------------------------------------------------
|
59 |
-
_LABELS = {
|
60 |
-
"pathogens": [
|
61 |
-
"O",
|
62 |
-
"B-COMM", "I-COMM",
|
63 |
-
"B-PROB", "I-PROB",
|
64 |
-
"B-OPPO", "I-OPPO",
|
65 |
-
"B-NEGA", "I-NEGA",
|
66 |
-
"B-STRA", "I-STRA",
|
67 |
-
"B-PATH", "I-PATH",
|
68 |
-
],
|
69 |
-
"strains": [
|
70 |
-
"O",
|
71 |
-
"B-STRA", "I-STRA",
|
72 |
-
"B-SPEC", "I-SPEC",
|
73 |
-
"B-NEGA", "I-NEGA",
|
74 |
-
],
|
75 |
-
}
|
76 |
-
|
77 |
-
# Map raw JSON tags (colour suffix stripped) to 4-letter codes
|
78 |
-
_CODE4 = {
|
79 |
-
"commensals": "COMM",
|
80 |
-
"probiotics": "PROB",
|
81 |
-
"opportunistics": "OPPO",
|
82 |
-
"negatives": "NEGA",
|
83 |
-
"strains": "STRA",
|
84 |
-
"pathogens": "PATH",
|
85 |
-
"species": "SPEC",
|
86 |
-
}
|
87 |
-
|
88 |
-
# ---------------------------------------------------------------------------
|
89 |
-
# DatasetBuilder
|
90 |
-
# ---------------------------------------------------------------------------
|
91 |
-
class TeaCurated(datasets.GeneratorBasedBuilder):
|
92 |
-
VERSION = datasets.Version("1.0.0")
|
93 |
-
|
94 |
-
BUILDER_CONFIGS = [
|
95 |
-
datasets.BuilderConfig(
|
96 |
-
name="pathogens",
|
97 |
-
version=VERSION,
|
98 |
-
description="Pathogen curation data",
|
99 |
-
),
|
100 |
-
datasets.BuilderConfig(
|
101 |
-
name="strains",
|
102 |
-
version=VERSION,
|
103 |
-
description="Strain curation data",
|
104 |
-
),
|
105 |
-
]
|
106 |
-
DEFAULT_CONFIG_NAME = "pathogens"
|
107 |
-
|
108 |
-
|
109 |
-
def _info(self):
|
110 |
-
return datasets.DatasetInfo(
|
111 |
-
description=_DESCRIPTION,
|
112 |
-
citation=_CITATION,
|
113 |
-
homepage=_HOMEPAGE,
|
114 |
-
license=_LICENSE,
|
115 |
-
features=datasets.Features(
|
116 |
-
{
|
117 |
-
"hash": datasets.Value("string"),
|
118 |
-
"tokens": datasets.Sequence(datasets.Value("string")),
|
119 |
-
"ner_tags": datasets.Sequence(
|
120 |
-
datasets.ClassLabel(names=_LABELS[self.config.name])
|
121 |
-
),
|
122 |
-
"locations": datasets.Value("string"),
|
123 |
-
"license": datasets.Value("string"),
|
124 |
-
"license_version": datasets.Value("string"),
|
125 |
-
"license_url": datasets.Value("string"),
|
126 |
-
"attribution_text": datasets.Value("string"),
|
127 |
-
}
|
128 |
-
),
|
129 |
-
supervised_keys=None,
|
130 |
-
)
|
131 |
-
|
132 |
-
|
133 |
-
def _split_generators(self, dl_manager):
|
134 |
-
extracted = Path(dl_manager.download_and_extract(_URL))
|
135 |
-
base_dir = extracted / "TEA_curated_data-1.1"
|
136 |
-
|
137 |
-
curation_json = (
|
138 |
-
base_dir / "curation_data" / self.config.name / f"{self.config.name}.json"
|
139 |
-
)
|
140 |
-
articles_dir = base_dir / "source_articles"
|
141 |
-
|
142 |
-
return [
|
143 |
-
datasets.SplitGenerator(
|
144 |
-
name=datasets.Split.TRAIN,
|
145 |
-
gen_kwargs={
|
146 |
-
"curation_path": curation_json,
|
147 |
-
"articles_dir": articles_dir,
|
148 |
-
},
|
149 |
-
)
|
150 |
-
]
|
151 |
-
|
152 |
-
|
153 |
-
def _generate_examples(self, curation_path: Path, articles_dir: Path):
|
154 |
-
with open(curation_path, encoding="utf-8") as f:
|
155 |
-
curation = json.load(f)
|
156 |
-
|
157 |
-
attribution = _parse_attribution_file(articles_dir.parent / "attribution.txt")
|
158 |
-
|
159 |
-
for art_hash, annotation in curation.items():
|
160 |
-
if art_hash not in attribution:
|
161 |
-
raise ValueError(f"Attribution missing for article hash {art_hash!r}")
|
162 |
-
|
163 |
-
txt_path = articles_dir / art_hash / f"{art_hash}.txt"
|
164 |
-
text = txt_path.read_text(encoding="utf-8")
|
165 |
-
text = re.sub(r"\$/i\$", "", text)
|
166 |
-
text = re.sub(r"\$i\$", "", text)
|
167 |
-
text = re.sub(r"\s+", " ", text)
|
168 |
-
|
169 |
-
tokens = text.split()
|
170 |
-
ner_tags = ["O"] * len(tokens)
|
171 |
-
|
172 |
-
for raw_tag, locs in annotation.items():
|
173 |
-
code = _CODE4[raw_tag.split("/")[0]]
|
174 |
-
for loc in locs: # e.g. "353+2"
|
175 |
-
start, length = map(int, loc.split("+"))
|
176 |
-
for i in range(start, start + length):
|
177 |
-
if i < len(tokens):
|
178 |
-
ner_tags[i] = ("B-" if i == start else "I-") + code
|
179 |
-
|
180 |
-
meta = attribution[art_hash]
|
181 |
-
|
182 |
-
yield art_hash, {
|
183 |
-
"hash": art_hash,
|
184 |
-
"tokens": tokens,
|
185 |
-
"ner_tags": ner_tags,
|
186 |
-
"locations": json.dumps(annotation),
|
187 |
-
"license": meta["license"],
|
188 |
-
"license_version": meta["license_version"],
|
189 |
-
"license_url": meta["license_url"],
|
190 |
-
"attribution_text": meta["text"],
|
191 |
-
}
|
192 |
-
|
193 |
-
# ---------------------------------------------------------------------------
|
194 |
-
# Attribution parser
|
195 |
-
# ---------------------------------------------------------------------------
|
196 |
-
def _parse_attribution_file(file_path: Path) -> dict[str, dict[str, str]]:
|
197 |
-
"""
|
198 |
-
Parse attribution.txt into
|
199 |
-
{checksum: {license, license_version, license_url, text}}.
|
200 |
-
"""
|
201 |
-
out = {}
|
202 |
-
csum = lic = lic_url = lic_ver = None
|
203 |
-
temp_lines = []
|
204 |
-
|
205 |
-
with open(file_path, encoding="utf-8") as fp:
|
206 |
-
for line in fp:
|
207 |
-
if "Checksum: " in line:
|
208 |
-
csum = line.split("Checksum: ")[1].strip()
|
209 |
-
elif "is obtained by the following licence: " in line:
|
210 |
-
lic_url = line.split("is obtained by the following licence: ")[1].strip()
|
211 |
-
lic_parts = (
|
212 |
-
line.split(
|
213 |
-
"is obtained by the following licence: http://creativecommons.org/licenses/"
|
214 |
-
)[1]
|
215 |
-
.strip()
|
216 |
-
.split("/")
|
217 |
-
)
|
218 |
-
lic_ver = lic_parts[1]
|
219 |
-
lic = f"CC-{lic_parts[0].upper()}"
|
220 |
-
out[csum] = {
|
221 |
-
"license": lic,
|
222 |
-
"license_version": lic_ver,
|
223 |
-
"license_url": lic_url,
|
224 |
-
}
|
225 |
-
else:
|
226 |
-
if len(line.strip()) > 0:
|
227 |
-
temp_lines.append(line.strip())
|
228 |
-
else:
|
229 |
-
# blank line marks end of a record
|
230 |
-
out[csum]["text"] = " ".join(temp_lines).strip()
|
231 |
-
csum = lic = lic_url = lic_ver = None
|
232 |
-
temp_lines = []
|
233 |
-
|
234 |
-
# flush last record if file does not end with newline
|
235 |
-
if csum and "text" not in out[csum]:
|
236 |
-
out[csum]["text"] = " ".join(temp_lines).strip()
|
237 |
-
|
238 |
-
return out
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|