Datasets:
Formats:
parquet
Sub-tasks:
language-modeling
Languages:
English
Size:
10M - 100M
Tags:
text-search
License:
Commit
•
b2cb1e0
1
Parent(s):
22bacfc
Delete loading script
Browse files- wiki_snippets.py +0 -210
wiki_snippets.py
DELETED
@@ -1,210 +0,0 @@
|
|
1 |
-
# WARNING: Please, do not use the code in this script as a template to create another script:
|
2 |
-
# - It is a bad practice to use `datasets.load_dataset` inside a loading script. Please, avoid doing it.
|
3 |
-
|
4 |
-
import json
|
5 |
-
import math
|
6 |
-
|
7 |
-
import datasets
|
8 |
-
|
9 |
-
|
10 |
-
logger = datasets.logging.get_logger(__name__)
|
11 |
-
|
12 |
-
|
13 |
-
_CITATION = """\
|
14 |
-
@ONLINE {wikidump,
|
15 |
-
author = {Wikimedia Foundation},
|
16 |
-
title = {Wikimedia Downloads},
|
17 |
-
url = {https://dumps.wikimedia.org}
|
18 |
-
}
|
19 |
-
"""
|
20 |
-
|
21 |
-
_DESCRIPTION = """\
|
22 |
-
Wikipedia version split into plain text snippets for dense semantic indexing.
|
23 |
-
"""
|
24 |
-
|
25 |
-
_LICENSE = (
|
26 |
-
"This work is licensed under the Creative Commons Attribution-ShareAlike "
|
27 |
-
"3.0 Unported License. To view a copy of this license, visit "
|
28 |
-
"http://creativecommons.org/licenses/by-sa/3.0/ or send a letter to "
|
29 |
-
"Creative Commons, PO Box 1866, Mountain View, CA 94042, USA."
|
30 |
-
)
|
31 |
-
|
32 |
-
|
33 |
-
def wiki40b_article_snippets(article, passage_len=100, overlap=0):
|
34 |
-
paragraphs = article["text"].split("\n")
|
35 |
-
aticle_idx = paragraphs.index("_START_ARTICLE_") + 1
|
36 |
-
article_title = paragraphs[aticle_idx] if aticle_idx < len(paragraphs) else ""
|
37 |
-
section_indices = [i + 1 for i, par in enumerate(paragraphs[:-1]) if par == "_START_SECTION_"]
|
38 |
-
par_tabs = [par.split(" ") for par in paragraphs]
|
39 |
-
word_map = [
|
40 |
-
(i, len(" ".join(par[:j])), w)
|
41 |
-
for i, par in enumerate(par_tabs)
|
42 |
-
if not par[0].startswith("_START_")
|
43 |
-
for j, w in enumerate(par)
|
44 |
-
if i > 0
|
45 |
-
]
|
46 |
-
step_size = passage_len - overlap
|
47 |
-
passages = []
|
48 |
-
for i in range(math.ceil(len(word_map) / step_size)):
|
49 |
-
pre_toks = word_map[i * step_size : i * step_size + passage_len]
|
50 |
-
start_section_id = max([0] + [j for j in section_indices if j <= pre_toks[0][0]])
|
51 |
-
section_ids = [j for j in section_indices if j >= start_section_id and j <= pre_toks[-1][0]]
|
52 |
-
section_ids = section_ids if len(section_ids) > 0 else [0]
|
53 |
-
passage_text = " ".join([w for p_id, s_id, w in pre_toks])
|
54 |
-
passages += [
|
55 |
-
{
|
56 |
-
"article_title": article_title,
|
57 |
-
"section_title": " & ".join([paragraphs[j] for j in section_ids]),
|
58 |
-
"wiki_id": article["wikidata_id"],
|
59 |
-
"start_paragraph": pre_toks[0][0],
|
60 |
-
"start_character": pre_toks[0][1],
|
61 |
-
"end_paragraph": pre_toks[-1][0],
|
62 |
-
"end_character": pre_toks[-1][1] + len(pre_toks[-1][2]) + 1,
|
63 |
-
"passage_text": passage_text.replace("_NEWLINE_", "\n"),
|
64 |
-
}
|
65 |
-
]
|
66 |
-
return passages
|
67 |
-
|
68 |
-
|
69 |
-
def wikipedia_article_snippets(article, passage_len=100, overlap=0):
|
70 |
-
paragraphs = [par for par in article["text"].split("\n") if not par.startswith("Category:")]
|
71 |
-
if "References" in paragraphs:
|
72 |
-
paragraphs = paragraphs[: paragraphs.index("References")]
|
73 |
-
article_title = article["title"]
|
74 |
-
section_indices = [
|
75 |
-
i + 1
|
76 |
-
for i, par in enumerate(paragraphs[:-2])
|
77 |
-
if paragraphs[i] == "" and paragraphs[i + 1] != "" and paragraphs[i + 2] != ""
|
78 |
-
]
|
79 |
-
par_tabs = [par.split(" ") for par in paragraphs]
|
80 |
-
word_map = [(i, len(" ".join(par[:j])), w) for i, par in enumerate(par_tabs) for j, w in enumerate(par)]
|
81 |
-
step_size = passage_len - overlap
|
82 |
-
passages = []
|
83 |
-
for i in range(math.ceil(len(word_map) / step_size)):
|
84 |
-
pre_toks = word_map[i * step_size : i * step_size + passage_len]
|
85 |
-
start_section_id = max([0] + [j for j in section_indices if j <= pre_toks[0][0]])
|
86 |
-
section_ids = [j for j in section_indices if start_section_id <= j <= pre_toks[-1][0]]
|
87 |
-
section_ids = section_ids if len(section_ids) > 0 else [-1]
|
88 |
-
passage_text = " ".join([w for p_id, s_id, w in pre_toks])
|
89 |
-
passages += [
|
90 |
-
{
|
91 |
-
"article_title": article_title,
|
92 |
-
"section_title": " & ".join(["Start" if j == -1 else paragraphs[j].strip() for j in section_ids]),
|
93 |
-
"wiki_id": article_title.replace(" ", "_"),
|
94 |
-
"start_paragraph": pre_toks[0][0],
|
95 |
-
"start_character": pre_toks[0][1],
|
96 |
-
"end_paragraph": pre_toks[-1][0],
|
97 |
-
"end_character": pre_toks[-1][1] + len(pre_toks[-1][2]) + 1,
|
98 |
-
"passage_text": passage_text,
|
99 |
-
}
|
100 |
-
]
|
101 |
-
return passages
|
102 |
-
|
103 |
-
|
104 |
-
_SPLIT_FUNCTION_MAP = {
|
105 |
-
"wikipedia": wikipedia_article_snippets,
|
106 |
-
"wiki40b": wiki40b_article_snippets,
|
107 |
-
}
|
108 |
-
|
109 |
-
|
110 |
-
def generate_snippets(wikipedia, split_function, passage_len=100, overlap=0):
|
111 |
-
for i, article in enumerate(wikipedia):
|
112 |
-
for doc in split_function(article, passage_len, overlap):
|
113 |
-
part_id = json.dumps(
|
114 |
-
{
|
115 |
-
"datasets_id": i,
|
116 |
-
"wiki_id": doc["wiki_id"],
|
117 |
-
"sp": doc["start_paragraph"],
|
118 |
-
"sc": doc["start_character"],
|
119 |
-
"ep": doc["end_paragraph"],
|
120 |
-
"ec": doc["end_character"],
|
121 |
-
}
|
122 |
-
)
|
123 |
-
doc["_id"] = part_id
|
124 |
-
doc["datasets_id"] = i
|
125 |
-
yield doc
|
126 |
-
|
127 |
-
|
128 |
-
class WikiSnippetsConfig(datasets.BuilderConfig):
|
129 |
-
"""BuilderConfig for WikiSnippets."""
|
130 |
-
|
131 |
-
def __init__(
|
132 |
-
self, wikipedia_name="wiki40b", wikipedia_version_name="en", snippets_length=100, snippets_overlap=0, **kwargs
|
133 |
-
):
|
134 |
-
"""BuilderConfig for WikiSnippets.
|
135 |
-
Args:
|
136 |
-
**kwargs: keyword arguments forwarded to super.
|
137 |
-
"""
|
138 |
-
super(WikiSnippetsConfig, self).__init__(**kwargs)
|
139 |
-
self.wikipedia_name = wikipedia_name
|
140 |
-
self.wikipedia_version_name = wikipedia_version_name
|
141 |
-
self.snippets_length = snippets_length
|
142 |
-
self.snippets_overlap = snippets_overlap
|
143 |
-
|
144 |
-
|
145 |
-
class WikiSnippets(datasets.GeneratorBasedBuilder):
|
146 |
-
BUILDER_CONFIG_CLASS = WikiSnippetsConfig
|
147 |
-
BUILDER_CONFIGS = [
|
148 |
-
WikiSnippetsConfig(
|
149 |
-
name="wiki40b_en_100_0",
|
150 |
-
version=datasets.Version("1.0.0"),
|
151 |
-
wikipedia_name="wiki40b",
|
152 |
-
wikipedia_version_name="en",
|
153 |
-
snippets_length=100,
|
154 |
-
snippets_overlap=0,
|
155 |
-
),
|
156 |
-
WikiSnippetsConfig(
|
157 |
-
name="wikipedia_en_100_0",
|
158 |
-
version=datasets.Version("2.0.0"),
|
159 |
-
wikipedia_name="wikipedia",
|
160 |
-
wikipedia_version_name="20220301.en",
|
161 |
-
snippets_length=100,
|
162 |
-
snippets_overlap=0,
|
163 |
-
),
|
164 |
-
]
|
165 |
-
|
166 |
-
test_dummy_data = False
|
167 |
-
|
168 |
-
def _info(self):
|
169 |
-
return datasets.DatasetInfo(
|
170 |
-
description=_DESCRIPTION,
|
171 |
-
features=datasets.Features(
|
172 |
-
{
|
173 |
-
"_id": datasets.Value("string"),
|
174 |
-
"datasets_id": datasets.Value("int32"),
|
175 |
-
"wiki_id": datasets.Value("string"),
|
176 |
-
"start_paragraph": datasets.Value("int32"),
|
177 |
-
"start_character": datasets.Value("int32"),
|
178 |
-
"end_paragraph": datasets.Value("int32"),
|
179 |
-
"end_character": datasets.Value("int32"),
|
180 |
-
"article_title": datasets.Value("string"),
|
181 |
-
"section_title": datasets.Value("string"),
|
182 |
-
"passage_text": datasets.Value("string"),
|
183 |
-
}
|
184 |
-
),
|
185 |
-
supervised_keys=None,
|
186 |
-
homepage="https://dumps.wikimedia.org",
|
187 |
-
citation=_CITATION,
|
188 |
-
)
|
189 |
-
|
190 |
-
def _split_generators(self, dl_manager):
|
191 |
-
# WARNING: It is a bad practice to use `datasets.load_dataset` inside a loading script. Please, avoid doing it.
|
192 |
-
wikipedia = datasets.load_dataset(
|
193 |
-
path=self.config.wikipedia_name,
|
194 |
-
name=self.config.wikipedia_version_name,
|
195 |
-
)
|
196 |
-
|
197 |
-
return [
|
198 |
-
datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"wikipedia": wikipedia}),
|
199 |
-
]
|
200 |
-
|
201 |
-
def _generate_examples(self, wikipedia):
|
202 |
-
logger.info(f"generating examples from = {self.config.wikipedia_name} {self.config.wikipedia_version_name}")
|
203 |
-
for split in wikipedia:
|
204 |
-
dset = wikipedia[split]
|
205 |
-
split_function = _SPLIT_FUNCTION_MAP[self.config.wikipedia_name]
|
206 |
-
for doc in generate_snippets(
|
207 |
-
dset, split_function, passage_len=self.config.snippets_length, overlap=self.config.snippets_overlap
|
208 |
-
):
|
209 |
-
id_ = doc["_id"]
|
210 |
-
yield id_, doc
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|