mathiascreutz
commited on
Commit
•
bba20ca
1
Parent(s):
ee6c327
Data loader complete and documented
Browse files- opusparcus.py +65 -24
opusparcus.py
CHANGED
@@ -1,5 +1,6 @@
|
|
1 |
# coding=utf-8
|
2 |
-
# Copyright
|
|
|
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.
|
@@ -13,7 +14,7 @@
|
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
|
16 |
-
"""
|
17 |
|
18 |
import csv
|
19 |
import json
|
@@ -24,22 +25,36 @@ import bz2
|
|
24 |
# Add BibTeX citation
|
25 |
|
26 |
_CITATION = """\
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
},
|
32 |
-
year={
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
}
|
34 |
"""
|
35 |
|
36 |
_DESCRIPTION = """\
|
37 |
-
|
|
|
|
|
|
|
38 |
"""
|
39 |
|
40 |
-
_HOMEPAGE = ""
|
41 |
|
42 |
-
_LICENSE = ""
|
43 |
|
44 |
# The HuggingFace dataset library doesn't host the datasets but only
|
45 |
# points to the original files. This can be an arbitrary nested
|
@@ -61,9 +76,10 @@ class OpusparcusConfig(datasets.BuilderConfig):
|
|
61 |
def __init__(self, lang=None, quality=100, **kwargs):
|
62 |
"""BuilderConfig for Wikipedia.
|
63 |
Args:
|
64 |
-
|
65 |
-
|
66 |
-
|
|
|
67 |
**kwargs: keyword arguments forwarded to super.
|
68 |
"""
|
69 |
super(OpusparcusConfig, self).__init__(
|
@@ -94,19 +110,44 @@ QUALITIES = [ 100, 95, 90, 85, 80, 75, 70, 65, 60 ]
|
|
94 |
|
95 |
class Opusparcus(datasets.GeneratorBasedBuilder):
|
96 |
|
97 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
|
99 |
-
# This is
|
100 |
-
# If you don't want/need to define several sub-sets in your dataset,
|
101 |
-
# just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.
|
102 |
-
# If you need to make complex sub-parts in the datasets with configurable options
|
103 |
-
# You can create your own builder configuration class to store attribute, inheriting from datasets.BuilderConfig
|
104 |
BUILDER_CONFIG_CLASS = OpusparcusConfig
|
105 |
-
|
106 |
-
#
|
107 |
-
# data = datasets.load_dataset('
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
108 |
BUILDER_CONFIGS = [
|
109 |
-
OpusparcusConfig(lang=lang, quality=quality, version=_VERSION)
|
|
|
110 |
]
|
111 |
|
112 |
# There is no default configuration. User always needs to specify one:
|
|
|
1 |
# coding=utf-8
|
2 |
+
# Copyright 2021 The HuggingFace Datasets Authors and
|
3 |
+
# the current dataset script contributor (Mathias Creutz).
|
4 |
#
|
5 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
6 |
# you may not use this file except in compliance with the License.
|
|
|
14 |
# See the License for the specific language governing permissions and
|
15 |
# limitations under the License.
|
16 |
|
17 |
+
"""Data loader for the Opusparcus paraphrase corpus."""
|
18 |
|
19 |
import csv
|
20 |
import json
|
|
|
25 |
# Add BibTeX citation
|
26 |
|
27 |
_CITATION = """\
|
28 |
+
@InProceedings{creutz:lrec2018,
|
29 |
+
title = {Open Subtitles Paraphrase Corpus for Six Languages},
|
30 |
+
author={Mathias Creutz},
|
31 |
+
booktitle={Proceedings of the 11th edition of the Language Resources
|
32 |
+
and Evaluation Conference (LREC 2018)},
|
33 |
+
year={2018},
|
34 |
+
month = {May 7-12},
|
35 |
+
address = {Miyazaki, Japan},
|
36 |
+
editor = {Nicoletta Calzolari (Conference chair) and Khalid Choukri
|
37 |
+
and Christopher Cieri and Thierry Declerck and Sara Goggi and Koiti
|
38 |
+
Hasida and Hitoshi Isahara and Bente Maegaard and Joseph Mariani and
|
39 |
+
Hélène Mazo and Asuncion Moreno and Jan Odijk and Stelios Piperidis
|
40 |
+
and Takenobu Tokunaga},
|
41 |
+
publisher = {European Language Resources Association (ELRA)},
|
42 |
+
isbn = {979-10-95546-00-9},
|
43 |
+
language = {english},
|
44 |
+
url={http://www.lrec-conf.org/proceedings/lrec2018/pdf/131.pdf}
|
45 |
}
|
46 |
"""
|
47 |
|
48 |
_DESCRIPTION = """\
|
49 |
+
Opusparcus is a paraphrase corpus for six European languages: German,
|
50 |
+
English, Finnish, French, Russian, and Swedish. The paraphrases are
|
51 |
+
extracted from the OpenSubtitles2016 corpus, which contains subtitles
|
52 |
+
from movies and TV shows.
|
53 |
"""
|
54 |
|
55 |
+
_HOMEPAGE = "http://urn.fi/urn:nbn:fi:lb-2018021221"
|
56 |
|
57 |
+
_LICENSE = "CC-BY-NC"
|
58 |
|
59 |
# The HuggingFace dataset library doesn't host the datasets but only
|
60 |
# points to the original files. This can be an arbitrary nested
|
|
|
76 |
def __init__(self, lang=None, quality=100, **kwargs):
|
77 |
"""BuilderConfig for Wikipedia.
|
78 |
Args:
|
79 |
+
lang: string, two letter language code:
|
80 |
+
de, en, fi, fr, ru, sv
|
81 |
+
quality: int, filter training set according to quality:
|
82 |
+
[ 60, 65, 70, 75, 80, 85, 90, 95, 100]
|
83 |
**kwargs: keyword arguments forwarded to super.
|
84 |
"""
|
85 |
super(OpusparcusConfig, self).__init__(
|
|
|
110 |
|
111 |
class Opusparcus(datasets.GeneratorBasedBuilder):
|
112 |
|
113 |
+
"""Opusparcus is a paraphrase corpus for six European languages:
|
114 |
+
German, English, Finnish, French, Russian, and Swedish. The
|
115 |
+
paraphrases are extracted from the OpenSubtitles2016 corpus, which
|
116 |
+
contains subtitles from movies and TV shows.
|
117 |
+
|
118 |
+
The data in Opusparcus has been extracted from OpenSubtitles2016
|
119 |
+
(http://opus.nlpl.eu/OpenSubtitles2016.php), which is in turn
|
120 |
+
based on data from http://www.opensubtitles.org/.
|
121 |
+
|
122 |
+
For each target language, the Opusparcus data have been
|
123 |
+
partitioned into three types of data sets: training, validation
|
124 |
+
and test sets. The training sets are large, consisting of millions
|
125 |
+
of sentence pairs, and have been compiled automatically, with the
|
126 |
+
help of probabilistic ranking functions. The development and test
|
127 |
+
sets consist of sentence pairs that have been annotated manually;
|
128 |
+
each set contains approximately 1000 sentence pairs that have been
|
129 |
+
verified to be acceptable paraphrases by two indepedent
|
130 |
+
annotators.
|
131 |
+
"""
|
132 |
|
133 |
+
# This is a dataset with multiple configurations.
|
|
|
|
|
|
|
|
|
134 |
BUILDER_CONFIG_CLASS = OpusparcusConfig
|
135 |
+
|
136 |
+
# You can load configurations as follows:
|
137 |
+
# data = datasets.load_dataset('GEM/opusparcus', lang='de')
|
138 |
+
# data = datasets.load_dataset('GEM/opusparcus', lang='fr', quality='75')
|
139 |
+
# etc.
|
140 |
+
#
|
141 |
+
# The language parameter is compulsory, whereas the quality
|
142 |
+
# parameter is not (the default value being 100).
|
143 |
+
#
|
144 |
+
# The above commands can alternatively be expressed as:
|
145 |
+
# data = datasets.load_dataset('GEM/opusparcus', 'de.100')
|
146 |
+
# data = datasets.load_dataset('GEM/opusparcus', 'fr.75')
|
147 |
+
|
148 |
BUILDER_CONFIGS = [
|
149 |
+
OpusparcusConfig(lang=lang, quality=quality, version=_VERSION) \
|
150 |
+
for lang in LANGS for quality in QUALITIES
|
151 |
]
|
152 |
|
153 |
# There is no default configuration. User always needs to specify one:
|