system HF staff commited on
Commit
cc97560
0 Parent(s):

Update files from the datasets library (from 1.0.0)

Browse files

Release notes: https://github.com/huggingface/datasets/releases/tag/1.0.0

.gitattributes ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bin.* filter=lfs diff=lfs merge=lfs -text
5
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.model filter=lfs diff=lfs merge=lfs -text
12
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
13
+ *.onnx filter=lfs diff=lfs merge=lfs -text
14
+ *.ot filter=lfs diff=lfs merge=lfs -text
15
+ *.parquet filter=lfs diff=lfs merge=lfs -text
16
+ *.pb filter=lfs diff=lfs merge=lfs -text
17
+ *.pt filter=lfs diff=lfs merge=lfs -text
18
+ *.pth filter=lfs diff=lfs merge=lfs -text
19
+ *.rar filter=lfs diff=lfs merge=lfs -text
20
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
21
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
22
+ *.tflite filter=lfs diff=lfs merge=lfs -text
23
+ *.tgz filter=lfs diff=lfs merge=lfs -text
24
+ *.xz filter=lfs diff=lfs merge=lfs -text
25
+ *.zip filter=lfs diff=lfs merge=lfs -text
26
+ *.zstandard filter=lfs diff=lfs merge=lfs -text
27
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
dataset_infos.json ADDED
@@ -0,0 +1 @@
 
1
+ {"cs-en": {"description": "Translate dataset based on the data from statmt.org.\n\nVersions exists for the different years using a combination of multiple data\nsources. The base `wmt_translate` allows you to create your own config to choose\nyour own data/language pair by creating a custom `datasets.translate.wmt.WmtConfig`.\n\n```\nconfig = datasets.wmt.WmtConfig(\n version=\"0.0.1\",\n language_pair=(\"fr\", \"de\"),\n subsets={\n datasets.Split.TRAIN: [\"commoncrawl_frde\"],\n datasets.Split.VALIDATION: [\"euelections_dev2019\"],\n },\n)\nbuilder = datasets.builder(\"wmt_translate\", config=config)\n```\n\n", "citation": "\n@InProceedings{bojar-EtAl:2015:WMT,\n author = {Bojar, Ond\u000b{r}ej and Chatterjee, Rajen and Federmann, Christian and Haddow, Barry and Huck, Matthias and Hokamp, Chris and Koehn, Philipp and Logacheva, Varvara and Monz, Christof and Negri, Matteo and Post, Matt and Scarton, Carolina and Specia, Lucia and Turchi, Marco},\n title = {Findings of the 2015 Workshop on Statistical Machine Translation},\n booktitle = {Proceedings of the Tenth Workshop on Statistical Machine Translation},\n month = {September},\n year = {2015},\n address = {Lisbon, Portugal},\n publisher = {Association for Computational Linguistics},\n pages = {1--46},\n url = {http://aclweb.org/anthology/W15-3001}\n}\n", "homepage": "http://www.statmt.org/wmt15/translation-task.html", "license": "", "features": {"translation": {"languages": ["cs", "en"], "id": null, "_type": "Translation"}}, "supervised_keys": {"input": "cs", "output": "en"}, "builder_name": "wmt15", "config_name": "cs-en", "version": {"version_str": "1.0.0", "description": null, "datasets_version_to_prepare": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"test": {"name": "test", "num_bytes": 572203, "num_examples": 2656, "dataset_name": "wmt15"}, "train": {"name": "train", "num_bytes": 283007221, "num_examples": 959768, "dataset_name": "wmt15"}, "validation": {"name": "validation", "num_bytes": 757817, "num_examples": 3003, "dataset_name": "wmt15"}}, "download_checksums": {"http://www.statmt.org/wmt13/training-parallel-europarl-v7.tgz": {"num_bytes": 657632379, "checksum": "0224c7c710c8a063dfd893b0cc0830202d61f4c75c17eb8e31836103d27d96e7"}, "http://www.statmt.org/wmt13/training-parallel-commoncrawl.tgz": {"num_bytes": 918311367, "checksum": "c7a74e2ea01ac6c920123108627e35278d4ccb5701e15428ffa34de86fa3a9e5"}, "http://www.statmt.org/wmt15/training-parallel-nc-v10.tgz": {"num_bytes": 125136590, "checksum": "11e053e7ea29e87e69a1ad05e61cb559820947ecec144ea8c1d6cf05017c9602"}, "http://data.statmt.org/wmt19/translation-task/dev.tgz": {"num_bytes": 38654961, "checksum": "7a7deccf82ebb05ba508dba5eb21356492224e8f630ec4f992132b029b4b25e7"}}, "download_size": 1739735297, "dataset_size": 284337241, "size_in_bytes": 2024072538}}
dummy/cs-en/1.0.0/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:71ce11fc038d3a1fdca3050fb0753134e670dc8f3967f1951199d56ed7021fc7
3
+ size 4624
wmt15.py ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
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
+ # Lint as: python3
17
+ """WMT15: Translate dataset."""
18
+
19
+ import datasets
20
+
21
+ from .wmt_utils import Wmt, WmtConfig
22
+
23
+
24
+ _URL = "http://www.statmt.org/wmt15/translation-task.html"
25
+ _CITATION = """
26
+ @InProceedings{bojar-EtAl:2015:WMT,
27
+ author = {Bojar, Ond\v{r}ej and Chatterjee, Rajen and Federmann, Christian and Haddow, Barry and Huck, Matthias and Hokamp, Chris and Koehn, Philipp and Logacheva, Varvara and Monz, Christof and Negri, Matteo and Post, Matt and Scarton, Carolina and Specia, Lucia and Turchi, Marco},
28
+ title = {Findings of the 2015 Workshop on Statistical Machine Translation},
29
+ booktitle = {Proceedings of the Tenth Workshop on Statistical Machine Translation},
30
+ month = {September},
31
+ year = {2015},
32
+ address = {Lisbon, Portugal},
33
+ publisher = {Association for Computational Linguistics},
34
+ pages = {1--46},
35
+ url = {http://aclweb.org/anthology/W15-3001}
36
+ }
37
+ """
38
+
39
+ _LANGUAGE_PAIRS = [(lang, "en") for lang in ["cs", "de", "fi", "fr", "ru"]]
40
+
41
+
42
+ class Wmt15(Wmt):
43
+ """WMT 15 translation datasets for all {xx, "en"} language pairs."""
44
+
45
+ BUILDER_CONFIGS = [
46
+ WmtConfig( # pylint:disable=g-complex-comprehension
47
+ description="WMT 2015 %s-%s translation task dataset." % (l1, l2),
48
+ url=_URL,
49
+ citation=_CITATION,
50
+ language_pair=(l1, l2),
51
+ version=datasets.Version("1.0.0"),
52
+ )
53
+ for l1, l2 in _LANGUAGE_PAIRS
54
+ ]
55
+
56
+ @property
57
+ def manual_download_instructions(self):
58
+ if self.config.language_pair[1] in ["cs", "hi", "ru"]:
59
+ return "Please download the data manually as explained. TODO(PVP)"
60
+
61
+ @property
62
+ def _subsets(self):
63
+ return {
64
+ datasets.Split.TRAIN: [
65
+ "europarl_v7",
66
+ "europarl_v8_16",
67
+ "commoncrawl",
68
+ "multiun",
69
+ "newscommentary_v10",
70
+ "gigafren",
71
+ "czeng_10",
72
+ "yandexcorpus",
73
+ "wikiheadlines_fi",
74
+ "wikiheadlines_ru",
75
+ ],
76
+ datasets.Split.VALIDATION: ["newsdev2015", "newsdiscussdev2015", "newstest2014"],
77
+ datasets.Split.TEST: ["newstest2015", "newsdiscusstest2015"],
78
+ }
wmt_utils.py ADDED
@@ -0,0 +1,1018 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
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
+ # Lint as: python3
17
+ """WMT: Translate dataset."""
18
+
19
+ from __future__ import absolute_import, division, print_function
20
+
21
+ import codecs
22
+ import functools
23
+ import glob
24
+ import gzip
25
+ import itertools
26
+ import logging
27
+ import os
28
+ import re
29
+ import xml.etree.cElementTree as ElementTree
30
+ from abc import ABC, abstractmethod
31
+
32
+ import six
33
+
34
+ import datasets
35
+
36
+
37
+ _DESCRIPTION = """\
38
+ Translate dataset based on the data from statmt.org.
39
+
40
+ Versions exists for the different years using a combination of multiple data
41
+ sources. The base `wmt_translate` allows you to create your own config to choose
42
+ your own data/language pair by creating a custom `datasets.translate.wmt.WmtConfig`.
43
+
44
+ ```
45
+ config = datasets.wmt.WmtConfig(
46
+ version="0.0.1",
47
+ language_pair=("fr", "de"),
48
+ subsets={
49
+ datasets.Split.TRAIN: ["commoncrawl_frde"],
50
+ datasets.Split.VALIDATION: ["euelections_dev2019"],
51
+ },
52
+ )
53
+ builder = datasets.builder("wmt_translate", config=config)
54
+ ```
55
+
56
+ """
57
+
58
+
59
+ CWMT_SUBSET_NAMES = ["casia2015", "casict2011", "casict2015", "datum2015", "datum2017", "neu2017"]
60
+
61
+
62
+ class SubDataset(object):
63
+ """Class to keep track of information on a sub-dataset of WMT."""
64
+
65
+ def __init__(self, name, target, sources, url, path, manual_dl_files=None):
66
+ """Sub-dataset of WMT.
67
+
68
+ Args:
69
+ name: `string`, a unique dataset identifier.
70
+ target: `string`, the target language code.
71
+ sources: `set<string>`, the set of source language codes.
72
+ url: `string` or `(string, string)`, URL(s) or URL template(s) specifying
73
+ where to download the raw data from. If two strings are provided, the
74
+ first is used for the source language and the second for the target.
75
+ Template strings can either contain '{src}' placeholders that will be
76
+ filled in with the source language code, '{0}' and '{1}' placeholders
77
+ that will be filled in with the source and target language codes in
78
+ alphabetical order, or all 3.
79
+ path: `string` or `(string, string)`, path(s) or path template(s)
80
+ specifing the path to the raw data relative to the root of the
81
+ downloaded archive. If two strings are provided, the dataset is assumed
82
+ to be made up of parallel text files, the first being the source and the
83
+ second the target. If one string is provided, both languages are assumed
84
+ to be stored within the same file and the extension is used to determine
85
+ how to parse it. Template strings should be formatted the same as in
86
+ `url`.
87
+ manual_dl_files: `<list>(string)` (optional), the list of files that must
88
+ be manually downloaded to the data directory.
89
+ """
90
+ self._paths = (path,) if isinstance(path, six.string_types) else path
91
+ self._urls = (url,) if isinstance(url, six.string_types) else url
92
+ self._manual_dl_files = manual_dl_files if manual_dl_files else []
93
+ self.name = name
94
+ self.target = target
95
+ self.sources = set(sources)
96
+
97
+ def _inject_language(self, src, strings):
98
+ """Injects languages into (potentially) template strings."""
99
+ if src not in self.sources:
100
+ raise ValueError("Invalid source for '{0}': {1}".format(self.name, src))
101
+
102
+ def _format_string(s):
103
+ if "{0}" in s and "{1}" and "{src}" in s:
104
+ return s.format(*sorted([src, self.target]), src=src)
105
+ elif "{0}" in s and "{1}" in s:
106
+ return s.format(*sorted([src, self.target]))
107
+ elif "{src}" in s:
108
+ return s.format(src=src)
109
+ else:
110
+ return s
111
+
112
+ return [_format_string(s) for s in strings]
113
+
114
+ def get_url(self, src):
115
+ return self._inject_language(src, self._urls)
116
+
117
+ def get_manual_dl_files(self, src):
118
+ return self._inject_language(src, self._manual_dl_files)
119
+
120
+ def get_path(self, src):
121
+ return self._inject_language(src, self._paths)
122
+
123
+
124
+ # Subsets used in the training sets for various years of WMT.
125
+ _TRAIN_SUBSETS = [
126
+ # pylint:disable=line-too-long
127
+ SubDataset(
128
+ name="commoncrawl",
129
+ target="en", # fr-de pair in commoncrawl_frde
130
+ sources={"cs", "de", "es", "fr", "ru"},
131
+ url="http://www.statmt.org/wmt13/training-parallel-commoncrawl.tgz",
132
+ path=("commoncrawl.{src}-en.{src}", "commoncrawl.{src}-en.en"),
133
+ ),
134
+ SubDataset(
135
+ name="commoncrawl_frde",
136
+ target="de",
137
+ sources={"fr"},
138
+ url=(
139
+ "http://data.statmt.org/wmt19/translation-task/fr-de/bitexts/commoncrawl.fr.gz",
140
+ "http://data.statmt.org/wmt19/translation-task/fr-de/bitexts/commoncrawl.de.gz",
141
+ ),
142
+ path=("", ""),
143
+ ),
144
+ SubDataset(
145
+ name="czeng_10",
146
+ target="en",
147
+ sources={"cs"},
148
+ url="http://ufal.mff.cuni.cz/czeng/czeng10",
149
+ manual_dl_files=["data-plaintext-format.%d.tar" % i for i in range(10)],
150
+ # Each tar contains multiple files, which we process specially in
151
+ # _parse_czeng.
152
+ path=("data.plaintext-format/??train.gz",) * 10,
153
+ ),
154
+ SubDataset(
155
+ name="czeng_16pre",
156
+ target="en",
157
+ sources={"cs"},
158
+ url="http://ufal.mff.cuni.cz/czeng/czeng16pre",
159
+ manual_dl_files=["czeng16pre.deduped-ignoring-sections.txt.gz"],
160
+ path="",
161
+ ),
162
+ SubDataset(
163
+ name="czeng_16",
164
+ target="en",
165
+ sources={"cs"},
166
+ url="http://ufal.mff.cuni.cz/czeng",
167
+ manual_dl_files=["data-plaintext-format.%d.tar" % i for i in range(10)],
168
+ # Each tar contains multiple files, which we process specially in
169
+ # _parse_czeng.
170
+ path=("data.plaintext-format/??train.gz",) * 10,
171
+ ),
172
+ SubDataset(
173
+ # This dataset differs from the above in the filtering that is applied
174
+ # during parsing.
175
+ name="czeng_17",
176
+ target="en",
177
+ sources={"cs"},
178
+ url="http://ufal.mff.cuni.cz/czeng",
179
+ manual_dl_files=["data-plaintext-format.%d.tar" % i for i in range(10)],
180
+ # Each tar contains multiple files, which we process specially in
181
+ # _parse_czeng.
182
+ path=("data.plaintext-format/??train.gz",) * 10,
183
+ ),
184
+ SubDataset(
185
+ name="dcep_v1",
186
+ target="en",
187
+ sources={"lv"},
188
+ url="http://data.statmt.org/wmt17/translation-task/dcep.lv-en.v1.tgz",
189
+ path=("dcep.en-lv/dcep.lv", "dcep.en-lv/dcep.en"),
190
+ ),
191
+ SubDataset(
192
+ name="europarl_v7",
193
+ target="en",
194
+ sources={"cs", "de", "es", "fr"},
195
+ url="http://www.statmt.org/wmt13/training-parallel-europarl-v7.tgz",
196
+ path=("training/europarl-v7.{src}-en.{src}", "training/europarl-v7.{src}-en.en"),
197
+ ),
198
+ SubDataset(
199
+ name="europarl_v7_frde",
200
+ target="de",
201
+ sources={"fr"},
202
+ url=(
203
+ "http://data.statmt.org/wmt19/translation-task/fr-de/bitexts/europarl-v7.fr.gz",
204
+ "http://data.statmt.org/wmt19/translation-task/fr-de/bitexts/europarl-v7.de.gz",
205
+ ),
206
+ path=("", ""),
207
+ ),
208
+ SubDataset(
209
+ name="europarl_v8_18",
210
+ target="en",
211
+ sources={"et", "fi"},
212
+ url="http://data.statmt.org/wmt18/translation-task/training-parallel-ep-v8.tgz",
213
+ path=("training/europarl-v8.{src}-en.{src}", "training/europarl-v8.{src}-en.en"),
214
+ ),
215
+ SubDataset(
216
+ name="europarl_v8_16",
217
+ target="en",
218
+ sources={"fi", "ro"},
219
+ url="http://data.statmt.org/wmt16/translation-task/training-parallel-ep-v8.tgz",
220
+ path=("training-parallel-ep-v8/europarl-v8.{src}-en.{src}", "training-parallel-ep-v8/europarl-v8.{src}-en.en"),
221
+ ),
222
+ SubDataset(
223
+ name="europarl_v9",
224
+ target="en",
225
+ sources={"cs", "de", "fi", "lt"},
226
+ url="http://www.statmt.org/europarl/v9/training/europarl-v9.{src}-en.tsv.gz",
227
+ path="",
228
+ ),
229
+ SubDataset(
230
+ name="gigafren",
231
+ target="en",
232
+ sources={"fr"},
233
+ url="http://www.statmt.org/wmt10/training-giga-fren.tar",
234
+ path=("giga-fren.release2.fixed.fr.gz", "giga-fren.release2.fixed.en.gz"),
235
+ ),
236
+ SubDataset(
237
+ name="hindencorp_01",
238
+ target="en",
239
+ sources={"hi"},
240
+ url="http://ufallab.ms.mff.cuni.cz/~bojar/hindencorp",
241
+ manual_dl_files=["hindencorp0.1.gz"],
242
+ path="",
243
+ ),
244
+ SubDataset(
245
+ name="leta_v1",
246
+ target="en",
247
+ sources={"lv"},
248
+ url="http://data.statmt.org/wmt17/translation-task/leta.v1.tgz",
249
+ path=("LETA-lv-en/leta.lv", "LETA-lv-en/leta.en"),
250
+ ),
251
+ SubDataset(
252
+ name="multiun",
253
+ target="en",
254
+ sources={"es", "fr"},
255
+ url="http://www.statmt.org/wmt13/training-parallel-un.tgz",
256
+ path=("un/undoc.2000.{src}-en.{src}", "un/undoc.2000.{src}-en.en"),
257
+ ),
258
+ SubDataset(
259
+ name="newscommentary_v9",
260
+ target="en",
261
+ sources={"cs", "de", "fr", "ru"},
262
+ url="http://www.statmt.org/wmt14/training-parallel-nc-v9.tgz",
263
+ path=("training/news-commentary-v9.{src}-en.{src}", "training/news-commentary-v9.{src}-en.en"),
264
+ ),
265
+ SubDataset(
266
+ name="newscommentary_v10",
267
+ target="en",
268
+ sources={"cs", "de", "fr", "ru"},
269
+ url="http://www.statmt.org/wmt15/training-parallel-nc-v10.tgz",
270
+ path=("news-commentary-v10.{src}-en.{src}", "news-commentary-v10.{src}-en.en"),
271
+ ),
272
+ SubDataset(
273
+ name="newscommentary_v11",
274
+ target="en",
275
+ sources={"cs", "de", "ru"},
276
+ url="http://data.statmt.org/wmt16/translation-task/training-parallel-nc-v11.tgz",
277
+ path=(
278
+ "training-parallel-nc-v11/news-commentary-v11.{src}-en.{src}",
279
+ "training-parallel-nc-v11/news-commentary-v11.{src}-en.en",
280
+ ),
281
+ ),
282
+ SubDataset(
283
+ name="newscommentary_v12",
284
+ target="en",
285
+ sources={"cs", "de", "ru", "zh"},
286
+ url="http://data.statmt.org/wmt17/translation-task/training-parallel-nc-v12.tgz",
287
+ path=("training/news-commentary-v12.{src}-en.{src}", "training/news-commentary-v12.{src}-en.en"),
288
+ ),
289
+ SubDataset(
290
+ name="newscommentary_v13",
291
+ target="en",
292
+ sources={"cs", "de", "ru", "zh"},
293
+ url="http://data.statmt.org/wmt18/translation-task/training-parallel-nc-v13.tgz",
294
+ path=(
295
+ "training-parallel-nc-v13/news-commentary-v13.{src}-en.{src}",
296
+ "training-parallel-nc-v13/news-commentary-v13.{src}-en.en",
297
+ ),
298
+ ),
299
+ SubDataset(
300
+ name="newscommentary_v14",
301
+ target="en", # fr-de pair in newscommentary_v14_frde
302
+ sources={"cs", "de", "kk", "ru", "zh"},
303
+ url="http://data.statmt.org/news-commentary/v14/training/news-commentary-v14.{0}-{1}.tsv.gz",
304
+ path="",
305
+ ),
306
+ SubDataset(
307
+ name="newscommentary_v14_frde",
308
+ target="de",
309
+ sources={"fr"},
310
+ url="http://data.statmt.org/news-commentary/v14/training/news-commentary-v14.de-fr.tsv.gz",
311
+ path="",
312
+ ),
313
+ SubDataset(
314
+ name="onlinebooks_v1",
315
+ target="en",
316
+ sources={"lv"},
317
+ url="http://data.statmt.org/wmt17/translation-task/books.lv-en.v1.tgz",
318
+ path=("farewell/farewell.lv", "farewell/farewell.en"),
319
+ ),
320
+ SubDataset(
321
+ name="paracrawl_v1",
322
+ target="en",
323
+ sources={"cs", "de", "et", "fi", "ru"},
324
+ url="https://s3.amazonaws.com/web-language-models/paracrawl/release1/paracrawl-release1.en-{src}.zipporah0-dedup-clean.tgz",
325
+ path=(
326
+ "paracrawl-release1.en-{src}.zipporah0-dedup-clean.{src}",
327
+ "paracrawl-release1.en-{src}.zipporah0-dedup-clean.en",
328
+ ),
329
+ ),
330
+ SubDataset(
331
+ name="paracrawl_v1_ru",
332
+ target="en",
333
+ sources={"ru"},
334
+ url="https://s3.amazonaws.com/web-language-models/paracrawl/release1/paracrawl-release1.en-ru.zipporah0-dedup-clean.tgz",
335
+ path=(
336
+ "paracrawl-release1.en-ru.zipporah0-dedup-clean.ru",
337
+ "paracrawl-release1.en-ru.zipporah0-dedup-clean.en",
338
+ ),
339
+ ),
340
+ SubDataset(
341
+ name="paracrawl_v3",
342
+ target="en", # fr-de pair in paracrawl_v3_frde
343
+ sources={"cs", "de", "fi", "lt"},
344
+ url="https://s3.amazonaws.com/web-language-models/paracrawl/release3/en-{src}.bicleaner07.tmx.gz",
345
+ path="",
346
+ ),
347
+ SubDataset(
348
+ name="paracrawl_v3_frde",
349
+ target="de",
350
+ sources={"fr"},
351
+ url=(
352
+ "http://data.statmt.org/wmt19/translation-task/fr-de/bitexts/de-fr.bicleaner07.de.gz",
353
+ "http://data.statmt.org/wmt19/translation-task/fr-de/bitexts/de-fr.bicleaner07.fr.gz",
354
+ ),
355
+ path=("", ""),
356
+ ),
357
+ SubDataset(
358
+ name="rapid_2016",
359
+ target="en",
360
+ sources={"de", "et", "fi"},
361
+ url="http://data.statmt.org/wmt18/translation-task/rapid2016.tgz",
362
+ path=("rapid2016.{0}-{1}.{src}", "rapid2016.{0}-{1}.en"),
363
+ ),
364
+ SubDataset(
365
+ name="rapid_2016_ltfi",
366
+ target="en",
367
+ sources={"fi", "lt"},
368
+ url="https://tilde-model.s3-eu-west-1.amazonaws.com/rapid2016.en-{src}.tmx.zip",
369
+ path="rapid2016.en-{src}.tmx",
370
+ ),
371
+ SubDataset(
372
+ name="rapid_2019",
373
+ target="en",
374
+ sources={"de"},
375
+ url="https://s3-eu-west-1.amazonaws.com/tilde-model/rapid2019.de-en.zip",
376
+ path=("rapid2019.de-en.de", "rapid2019.de-en.en"),
377
+ ),
378
+ SubDataset(
379
+ name="setimes_2",
380
+ target="en",
381
+ sources={"ro", "tr"},
382
+ url="http://opus.nlpl.eu/download.php?f=SETIMES/v2/tmx/en-{src}.tmx.gz",
383
+ path="",
384
+ ),
385
+ SubDataset(
386
+ name="uncorpus_v1",
387
+ target="en",
388
+ sources={"ru", "zh"},
389
+ url="https://storage.googleapis.com/tfdataset-data/downloadataset/uncorpus/UNv1.0.en-{src}.tar.gz",
390
+ path=("en-{src}/UNv1.0.en-{src}.{src}", "en-{src}/UNv1.0.en-{src}.en"),
391
+ ),
392
+ SubDataset(
393
+ name="wikiheadlines_fi",
394
+ target="en",
395
+ sources={"fi"},
396
+ url="http://www.statmt.org/wmt15/wiki-titles.tgz",
397
+ path="wiki/fi-en/titles.fi-en",
398
+ ),
399
+ SubDataset(
400
+ name="wikiheadlines_hi",
401
+ target="en",
402
+ sources={"hi"},
403
+ url="http://www.statmt.org/wmt14/wiki-titles.tgz",
404
+ path="wiki/hi-en/wiki-titles.hi-en",
405
+ ),
406
+ SubDataset(
407
+ # Verified that wmt14 and wmt15 files are identical.
408
+ name="wikiheadlines_ru",
409
+ target="en",
410
+ sources={"ru"},
411
+ url="http://www.statmt.org/wmt15/wiki-titles.tgz",
412
+ path="wiki/ru-en/wiki.ru-en",
413
+ ),
414
+ SubDataset(
415
+ name="wikititles_v1",
416
+ target="en",
417
+ sources={"cs", "de", "fi", "gu", "kk", "lt", "ru", "zh"},
418
+ url="http://data.statmt.org/wikititles/v1/wikititles-v1.{src}-en.tsv.gz",
419
+ path="",
420
+ ),
421
+ SubDataset(
422
+ name="yandexcorpus",
423
+ target="en",
424
+ sources={"ru"},
425
+ url="https://translate.yandex.ru/corpus?lang=en",
426
+ manual_dl_files=["1mcorpus.zip"],
427
+ path=("corpus.en_ru.1m.ru", "corpus.en_ru.1m.en"),
428
+ ),
429
+ # pylint:enable=line-too-long
430
+ ] + [
431
+ SubDataset( # pylint:disable=g-complex-comprehension
432
+ name=ss,
433
+ target="en",
434
+ sources={"zh"},
435
+ url="ftp://cwmt-wmt:cwmt-wmt@datasets.nju.edu.cn/parallel/%s.zip" % ss,
436
+ path=("%s/*_c[hn].txt" % ss, "%s/*_en.txt" % ss),
437
+ )
438
+ for ss in CWMT_SUBSET_NAMES
439
+ ]
440
+
441
+ _DEV_SUBSETS = [
442
+ SubDataset(
443
+ name="euelections_dev2019",
444
+ target="de",
445
+ sources={"fr"},
446
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
447
+ path=("dev/euelections_dev2019.fr-de.src.fr", "dev/euelections_dev2019.fr-de.tgt.de"),
448
+ ),
449
+ SubDataset(
450
+ name="newsdev2014",
451
+ target="en",
452
+ sources={"hi"},
453
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
454
+ path=("dev/newsdev2014.hi", "dev/newsdev2014.en"),
455
+ ),
456
+ SubDataset(
457
+ name="newsdev2015",
458
+ target="en",
459
+ sources={"fi"},
460
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
461
+ path=("dev/newsdev2015-fien-src.{src}.sgm", "dev/newsdev2015-fien-ref.en.sgm"),
462
+ ),
463
+ SubDataset(
464
+ name="newsdiscussdev2015",
465
+ target="en",
466
+ sources={"ro", "tr"},
467
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
468
+ path=("dev/newsdiscussdev2015-{src}en-src.{src}.sgm", "dev/newsdiscussdev2015-{src}en-ref.en.sgm"),
469
+ ),
470
+ SubDataset(
471
+ name="newsdev2016",
472
+ target="en",
473
+ sources={"ro", "tr"},
474
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
475
+ path=("dev/newsdev2016-{src}en-src.{src}.sgm", "dev/newsdev2016-{src}en-ref.en.sgm"),
476
+ ),
477
+ SubDataset(
478
+ name="newsdev2017",
479
+ target="en",
480
+ sources={"lv", "zh"},
481
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
482
+ path=("dev/newsdev2017-{src}en-src.{src}.sgm", "dev/newsdev2017-{src}en-ref.en.sgm"),
483
+ ),
484
+ SubDataset(
485
+ name="newsdev2018",
486
+ target="en",
487
+ sources={"et"},
488
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
489
+ path=("dev/newsdev2018-{src}en-src.{src}.sgm", "dev/newsdev2018-{src}en-ref.en.sgm"),
490
+ ),
491
+ SubDataset(
492
+ name="newsdev2019",
493
+ target="en",
494
+ sources={"gu", "kk", "lt"},
495
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
496
+ path=("dev/newsdev2019-{src}en-src.{src}.sgm", "dev/newsdev2019-{src}en-ref.en.sgm"),
497
+ ),
498
+ SubDataset(
499
+ name="newsdiscussdev2015",
500
+ target="en",
501
+ sources={"fr"},
502
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
503
+ path=("dev/newsdiscussdev2015-{src}en-src.{src}.sgm", "dev/newsdiscussdev2015-{src}en-ref.en.sgm"),
504
+ ),
505
+ SubDataset(
506
+ name="newsdiscusstest2015",
507
+ target="en",
508
+ sources={"fr"},
509
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
510
+ path=("dev/newsdiscusstest2015-{src}en-src.{src}.sgm", "dev/newsdiscusstest2015-{src}en-ref.en.sgm"),
511
+ ),
512
+ SubDataset(
513
+ name="newssyscomb2009",
514
+ target="en",
515
+ sources={"cs", "de", "es", "fr"},
516
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
517
+ path=("dev/newssyscomb2009.{src}", "dev/newssyscomb2009.en"),
518
+ ),
519
+ SubDataset(
520
+ name="newstest2008",
521
+ target="en",
522
+ sources={"cs", "de", "es", "fr", "hu"},
523
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
524
+ path=("dev/news-test2008.{src}", "dev/news-test2008.en"),
525
+ ),
526
+ SubDataset(
527
+ name="newstest2009",
528
+ target="en",
529
+ sources={"cs", "de", "es", "fr"},
530
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
531
+ path=("dev/newstest2009.{src}", "dev/newstest2009.en"),
532
+ ),
533
+ SubDataset(
534
+ name="newstest2010",
535
+ target="en",
536
+ sources={"cs", "de", "es", "fr"},
537
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
538
+ path=("dev/newstest2010.{src}", "dev/newstest2010.en"),
539
+ ),
540
+ SubDataset(
541
+ name="newstest2011",
542
+ target="en",
543
+ sources={"cs", "de", "es", "fr"},
544
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
545
+ path=("dev/newstest2011.{src}", "dev/newstest2011.en"),
546
+ ),
547
+ SubDataset(
548
+ name="newstest2012",
549
+ target="en",
550
+ sources={"cs", "de", "es", "fr", "ru"},
551
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
552
+ path=("dev/newstest2012.{src}", "dev/newstest2012.en"),
553
+ ),
554
+ SubDataset(
555
+ name="newstest2013",
556
+ target="en",
557
+ sources={"cs", "de", "es", "fr", "ru"},
558
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
559
+ path=("dev/newstest2013.{src}", "dev/newstest2013.en"),
560
+ ),
561
+ SubDataset(
562
+ name="newstest2014",
563
+ target="en",
564
+ sources={"cs", "de", "es", "fr", "hi", "ru"},
565
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
566
+ path=("dev/newstest2014-{src}en-src.{src}.sgm", "dev/newstest2014-{src}en-ref.en.sgm"),
567
+ ),
568
+ SubDataset(
569
+ name="newstest2015",
570
+ target="en",
571
+ sources={"cs", "de", "fi", "ru"},
572
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
573
+ path=("dev/newstest2015-{src}en-src.{src}.sgm", "dev/newstest2015-{src}en-ref.en.sgm"),
574
+ ),
575
+ SubDataset(
576
+ name="newsdiscusstest2015",
577
+ target="en",
578
+ sources={"fr"},
579
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
580
+ path=("dev/newsdiscusstest2015-{src}en-src.{src}.sgm", "dev/newsdiscusstest2015-{src}en-ref.en.sgm"),
581
+ ),
582
+ SubDataset(
583
+ name="newstest2016",
584
+ target="en",
585
+ sources={"cs", "de", "fi", "ro", "ru", "tr"},
586
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
587
+ path=("dev/newstest2016-{src}en-src.{src}.sgm", "dev/newstest2016-{src}en-ref.en.sgm"),
588
+ ),
589
+ SubDataset(
590
+ name="newstestB2016",
591
+ target="en",
592
+ sources={"fi"},
593
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
594
+ path=("dev/newstestB2016-enfi-ref.{src}.sgm", "dev/newstestB2016-enfi-src.en.sgm"),
595
+ ),
596
+ SubDataset(
597
+ name="newstest2017",
598
+ target="en",
599
+ sources={"cs", "de", "fi", "lv", "ru", "tr", "zh"},
600
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
601
+ path=("dev/newstest2017-{src}en-src.{src}.sgm", "dev/newstest2017-{src}en-ref.en.sgm"),
602
+ ),
603
+ SubDataset(
604
+ name="newstestB2017",
605
+ target="en",
606
+ sources={"fi"},
607
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
608
+ path=("dev/newstestB2017-fien-src.fi.sgm", "dev/newstestB2017-fien-ref.en.sgm"),
609
+ ),
610
+ SubDataset(
611
+ name="newstest2018",
612
+ target="en",
613
+ sources={"cs", "de", "et", "fi", "ru", "tr", "zh"},
614
+ url="http://data.statmt.org/wmt19/translation-task/dev.tgz",
615
+ path=("dev/newstest2018-{src}en-src.{src}.sgm", "dev/newstest2018-{src}en-ref.en.sgm"),
616
+ ),
617
+ ]
618
+
619
+ DATASET_MAP = {dataset.name: dataset for dataset in _TRAIN_SUBSETS + _DEV_SUBSETS}
620
+
621
+ _CZENG17_FILTER = SubDataset(
622
+ name="czeng17_filter",
623
+ target="en",
624
+ sources={"cs"},
625
+ url="http://ufal.mff.cuni.cz/czeng/download.php?f=convert_czeng16_to_17.pl.zip",
626
+ path="convert_czeng16_to_17.pl",
627
+ )
628
+
629
+
630
+ class WmtConfig(datasets.BuilderConfig):
631
+ """BuilderConfig for WMT."""
632
+
633
+ def __init__(self, url=None, citation=None, description=None, language_pair=(None, None), subsets=None, **kwargs):
634
+ """BuilderConfig for WMT.
635
+
636
+ Args:
637
+ url: The reference URL for the dataset.
638
+ citation: The paper citation for the dataset.
639
+ description: The description of the dataset.
640
+ language_pair: pair of languages that will be used for translation. Should
641
+ contain 2 letter coded strings. For example: ("en", "de").
642
+ configuration for the `datasets.features.text.TextEncoder` used for the
643
+ `datasets.features.text.Translation` features.
644
+ subsets: Dict[split, list[str]]. List of the subset to use for each of the
645
+ split. Note that WMT subclasses overwrite this parameter.
646
+ **kwargs: keyword arguments forwarded to super.
647
+ """
648
+ name = "%s-%s" % (language_pair[0], language_pair[1])
649
+ if "name" in kwargs: # Add name suffix for custom configs
650
+ name += "." + kwargs.pop("name")
651
+
652
+ super(WmtConfig, self).__init__(name=name, description=description, **kwargs)
653
+
654
+ self.url = url or "http://www.statmt.org"
655
+ self.citation = citation
656
+ self.language_pair = language_pair
657
+ self.subsets = subsets
658
+
659
+ # TODO(PVP): remove when manual dir works
660
+ # +++++++++++++++++++++
661
+ if language_pair[1] in ["cs", "hi", "ru"]:
662
+ assert NotImplementedError(
663
+ "The dataset for {}-en is currently not fully supported.".format(language_pair[1])
664
+ )
665
+ # +++++++++++++++++++++
666
+
667
+
668
+ class Wmt(ABC, datasets.GeneratorBasedBuilder):
669
+ """WMT translation dataset."""
670
+
671
+ def __init__(self, *args, **kwargs):
672
+ if type(self) == Wmt and "config" not in kwargs: # pylint: disable=unidiomatic-typecheck
673
+ raise ValueError(
674
+ "The raw `wmt_translate` can only be instantiated with the config "
675
+ "kwargs. You may want to use one of the `wmtYY_translate` "
676
+ "implementation instead to get the WMT dataset for a specific year."
677
+ )
678
+ super(Wmt, self).__init__(*args, **kwargs)
679
+
680
+ @property
681
+ @abstractmethod
682
+ def _subsets(self):
683
+ """Subsets that make up each split of the dataset."""
684
+ raise NotImplementedError("This is a abstract method")
685
+
686
+ @property
687
+ def subsets(self):
688
+ """Subsets that make up each split of the dataset for the language pair."""
689
+ source, target = self.config.language_pair
690
+ filtered_subsets = {}
691
+ for split, ss_names in self._subsets.items():
692
+ filtered_subsets[split] = []
693
+ for ss_name in ss_names:
694
+ dataset = DATASET_MAP[ss_name]
695
+ if dataset.target != target or source not in dataset.sources:
696
+ logging.info("Skipping sub-dataset that does not include language pair: %s", ss_name)
697
+ else:
698
+ filtered_subsets[split].append(ss_name)
699
+ logging.info("Using sub-datasets: %s", filtered_subsets)
700
+ return filtered_subsets
701
+
702
+ def _info(self):
703
+ src, target = self.config.language_pair
704
+ return datasets.DatasetInfo(
705
+ description=_DESCRIPTION,
706
+ features=datasets.Features(
707
+ {"translation": datasets.features.Translation(languages=self.config.language_pair)}
708
+ ),
709
+ supervised_keys=(src, target),
710
+ homepage=self.config.url,
711
+ citation=self.config.citation,
712
+ )
713
+
714
+ def _vocab_text_gen(self, split_subsets, extraction_map, language):
715
+ for _, ex in self._generate_examples(split_subsets, extraction_map, with_translation=False):
716
+ yield ex[language]
717
+
718
+ def _split_generators(self, dl_manager):
719
+ source, _ = self.config.language_pair
720
+ manual_paths_dict = {}
721
+ urls_to_download = {}
722
+ for ss_name in itertools.chain.from_iterable(self.subsets.values()):
723
+ if ss_name == "czeng_17":
724
+ # CzEng1.7 is CzEng1.6 with some blocks filtered out. We must download
725
+ # the filtering script so we can parse out which blocks need to be
726
+ # removed.
727
+ urls_to_download[_CZENG17_FILTER.name] = _CZENG17_FILTER.get_url(source)
728
+
729
+ # get dataset
730
+ dataset = DATASET_MAP[ss_name]
731
+ if dataset.get_manual_dl_files(source):
732
+ # TODO(PVP): following two lines skip configs that are incomplete for now
733
+ # +++++++++++++++++++++
734
+ logging.info("Skipping {} for now. Incomplete dataset for {}".format(dataset.name, self.config.name))
735
+ continue
736
+ # +++++++++++++++++++++
737
+
738
+ manual_dl_files = dataset.get_manual_dl_files(source)
739
+ manual_paths = [
740
+ os.path.join(os.path.abspath(os.path.expanduser(dl_manager.manual_dir)), fname)
741
+ for fname in manual_dl_files
742
+ ]
743
+ assert all(
744
+ os.path.exists(path) for path in manual_paths
745
+ ), "For {0}, you must manually download the following file(s) from {1} and place them in {2}: {3}".format(
746
+ dataset.name, dataset.get_url(source), dl_manager.manual_dir, ", ".join(manual_dl_files)
747
+ )
748
+
749
+ # set manual path for correct subset
750
+ manual_paths_dict[ss_name] = manual_paths
751
+ else:
752
+ urls_to_download[ss_name] = dataset.get_url(source)
753
+
754
+ # Download and extract files from URLs.
755
+ downloaded_files = dl_manager.download_and_extract(urls_to_download)
756
+ # Extract manually downloaded files.
757
+ manual_files = dl_manager.extract(manual_paths_dict)
758
+ extraction_map = dict(downloaded_files, **manual_files)
759
+
760
+ for language in self.config.language_pair:
761
+ self._vocab_text_gen(self.subsets[datasets.Split.TRAIN], extraction_map, language)
762
+
763
+ return [
764
+ datasets.SplitGenerator( # pylint:disable=g-complex-comprehension
765
+ name=split, gen_kwargs={"split_subsets": split_subsets, "extraction_map": extraction_map}
766
+ )
767
+ for split, split_subsets in self.subsets.items()
768
+ ]
769
+
770
+ def _generate_examples(self, split_subsets, extraction_map, with_translation=True):
771
+ """Returns the examples in the raw (text) form."""
772
+ source, _ = self.config.language_pair
773
+
774
+ def _get_local_paths(dataset, extract_dirs):
775
+ rel_paths = dataset.get_path(source)
776
+ if len(extract_dirs) == 1:
777
+ extract_dirs = extract_dirs * len(rel_paths)
778
+ return [
779
+ os.path.join(ex_dir, rel_path) if rel_path else ex_dir
780
+ for ex_dir, rel_path in zip(extract_dirs, rel_paths)
781
+ ]
782
+
783
+ for ss_name in split_subsets:
784
+ # TODO(PVP) remove following five lines when manual data works
785
+ # +++++++++++++++++++++
786
+ dataset = DATASET_MAP[ss_name]
787
+ source, _ = self.config.language_pair
788
+ if dataset.get_manual_dl_files(source):
789
+ logging.info("Skipping {} for now. Incomplete dataset for {}".format(dataset.name, self.config.name))
790
+ continue
791
+ # +++++++++++++++++++++
792
+
793
+ logging.info("Generating examples from: %s", ss_name)
794
+ dataset = DATASET_MAP[ss_name]
795
+ extract_dirs = extraction_map[ss_name]
796
+ files = _get_local_paths(dataset, extract_dirs)
797
+
798
+ if ss_name.startswith("czeng"):
799
+ if ss_name.endswith("16pre"):
800
+ sub_generator = functools.partial(_parse_tsv, language_pair=("en", "cs"))
801
+ elif ss_name.endswith("17"):
802
+ filter_path = _get_local_paths(_CZENG17_FILTER, extraction_map[_CZENG17_FILTER.name])[0]
803
+ sub_generator = functools.partial(_parse_czeng, filter_path=filter_path)
804
+ else:
805
+ sub_generator = _parse_czeng
806
+ elif ss_name == "hindencorp_01":
807
+ sub_generator = _parse_hindencorp
808
+ elif len(files) == 2:
809
+ if ss_name.endswith("_frde"):
810
+ sub_generator = _parse_frde_bitext
811
+ else:
812
+ sub_generator = _parse_parallel_sentences
813
+ elif len(files) == 1:
814
+ fname = files[0]
815
+ # Note: Due to formatting used by `download_manager`, the file
816
+ # extension may not be at the end of the file path.
817
+ if ".tsv" in fname:
818
+ sub_generator = _parse_tsv
819
+ elif (
820
+ ss_name.startswith("newscommentary_v14")
821
+ or ss_name.startswith("europarl_v9")
822
+ or ss_name.startswith("wikititles_v1")
823
+ ):
824
+ sub_generator = functools.partial(_parse_tsv, language_pair=self.config.language_pair)
825
+ elif "tmx" in fname or ss_name.startswith("paracrawl_v3"):
826
+ sub_generator = _parse_tmx
827
+ elif ss_name.startswith("wikiheadlines"):
828
+ sub_generator = _parse_wikiheadlines
829
+ else:
830
+ raise ValueError("Unsupported file format: %s" % fname)
831
+ else:
832
+ raise ValueError("Invalid number of files: %d" % len(files))
833
+
834
+ for sub_key, ex in sub_generator(*files):
835
+ if not all(ex.values()):
836
+ continue
837
+ # TODO(adarob): Add subset feature.
838
+ # ex["subset"] = subset
839
+ key = "{}/{}".format(ss_name, sub_key)
840
+ if with_translation is True:
841
+ ex = {"translation": ex}
842
+ yield key, ex
843
+
844
+
845
+ def _parse_parallel_sentences(f1, f2):
846
+ """Returns examples from parallel SGML or text files, which may be gzipped."""
847
+
848
+ def _parse_text(path):
849
+ """Returns the sentences from a single text file, which may be gzipped."""
850
+ split_path = path.split(".")
851
+
852
+ if split_path[-1] == "gz":
853
+ lang = split_path[-2]
854
+ with open(path, "rb") as f, gzip.GzipFile(fileobj=f) as g:
855
+ return g.read().decode("utf-8").split("\n"), lang
856
+
857
+ if split_path[-1] == "txt":
858
+ # CWMT
859
+ lang = split_path[-2].split("_")[-1]
860
+ lang = "zh" if lang in ("ch", "cn") else lang
861
+ else:
862
+ lang = split_path[-1]
863
+ with open(path, "rb") as f:
864
+ return f.read().decode("utf-8").split("\n"), lang
865
+
866
+ def _parse_sgm(path):
867
+ """Returns sentences from a single SGML file."""
868
+ lang = path.split(".")[-2]
869
+ sentences = []
870
+ # Note: We can't use the XML parser since some of the files are badly
871
+ # formatted.
872
+ seg_re = re.compile(r"<seg id=\"\d+\">(.*)</seg>")
873
+ with open(path, encoding="utf-8") as f:
874
+ for line in f:
875
+ seg_match = re.match(seg_re, line)
876
+ if seg_match:
877
+ assert len(seg_match.groups()) == 1
878
+ sentences.append(seg_match.groups()[0])
879
+ return sentences, lang
880
+
881
+ parse_file = _parse_sgm if f1.endswith(".sgm") else _parse_text
882
+
883
+ # Some datasets (e.g., CWMT) contain multiple parallel files specified with
884
+ # a wildcard. We sort both sets to align them and parse them one by one.
885
+ f1_files = sorted(glob.glob(f1))
886
+ f2_files = sorted(glob.glob(f2))
887
+
888
+ assert f1_files and f2_files, "No matching files found: %s, %s." % (f1, f2)
889
+ assert len(f1_files) == len(f2_files), "Number of files do not match: %d vs %d for %s vs %s." % (
890
+ len(f1_files),
891
+ len(f2_files),
892
+ f1,
893
+ f2,
894
+ )
895
+
896
+ for f_id, (f1_i, f2_i) in enumerate(zip(sorted(f1_files), sorted(f2_files))):
897
+ l1_sentences, l1 = parse_file(f1_i)
898
+ l2_sentences, l2 = parse_file(f2_i)
899
+
900
+ assert len(l1_sentences) == len(l2_sentences), "Sizes do not match: %d vs %d for %s vs %s." % (
901
+ len(l1_sentences),
902
+ len(l2_sentences),
903
+ f1_i,
904
+ f2_i,
905
+ )
906
+
907
+ for line_id, (s1, s2) in enumerate(zip(l1_sentences, l2_sentences)):
908
+ key = "{}/{}".format(f_id, line_id)
909
+ yield key, {l1: s1, l2: s2}
910
+
911
+
912
+ def _parse_frde_bitext(fr_path, de_path):
913
+ with open(fr_path, encoding="utf-8") as f:
914
+ fr_sentences = f.read().split("\n")
915
+ with open(de_path, encoding="utf-8") as f:
916
+ de_sentences = f.read().split("\n")
917
+ assert len(fr_sentences) == len(de_sentences), "Sizes do not match: %d vs %d for %s vs %s." % (
918
+ len(fr_sentences),
919
+ len(de_sentences),
920
+ fr_path,
921
+ de_path,
922
+ )
923
+ for line_id, (s1, s2) in enumerate(zip(fr_sentences, de_sentences)):
924
+ yield line_id, {"fr": s1, "de": s2}
925
+
926
+
927
+ def _parse_tmx(path):
928
+ """Generates examples from TMX file."""
929
+
930
+ def _get_tuv_lang(tuv):
931
+ for k, v in tuv.items():
932
+ if k.endswith("}lang"):
933
+ return v
934
+ raise AssertionError("Language not found in `tuv` attributes.")
935
+
936
+ def _get_tuv_seg(tuv):
937
+ segs = tuv.findall("seg")
938
+ assert len(segs) == 1, "Invalid number of segments: %d" % len(segs)
939
+ return segs[0].text
940
+
941
+ with open(path, "rb") as f:
942
+ if six.PY3:
943
+ # Workaround due to: https://github.com/tensorflow/tensorflow/issues/33563
944
+ utf_f = codecs.getreader("utf-8")(f)
945
+ else:
946
+ utf_f = f
947
+ for line_id, (_, elem) in enumerate(ElementTree.iterparse(utf_f)):
948
+ if elem.tag == "tu":
949
+ yield line_id, {_get_tuv_lang(tuv): _get_tuv_seg(tuv) for tuv in elem.iterfind("tuv")}
950
+ elem.clear()
951
+
952
+
953
+ def _parse_tsv(path, language_pair=None):
954
+ """Generates examples from TSV file."""
955
+ if language_pair is None:
956
+ lang_match = re.match(r".*\.([a-z][a-z])-([a-z][a-z])\.tsv", path)
957
+ assert lang_match is not None, "Invalid TSV filename: %s" % path
958
+ l1, l2 = lang_match.groups()
959
+ else:
960
+ l1, l2 = language_pair
961
+ with open(path, encoding="utf-8") as f:
962
+ for j, line in enumerate(f):
963
+ cols = line.split("\t")
964
+ if len(cols) != 2:
965
+ logging.warning("Skipping line %d in TSV (%s) with %d != 2 columns.", j, path, len(cols))
966
+ continue
967
+ s1, s2 = cols
968
+ yield j, {l1: s1.strip(), l2: s2.strip()}
969
+
970
+
971
+ def _parse_wikiheadlines(path):
972
+ """Generates examples from Wikiheadlines dataset file."""
973
+ lang_match = re.match(r".*\.([a-z][a-z])-([a-z][a-z])$", path)
974
+ assert lang_match is not None, "Invalid Wikiheadlines filename: %s" % path
975
+ l1, l2 = lang_match.groups()
976
+ with open(path, encoding="utf-8") as f:
977
+ for line_id, line in enumerate(f):
978
+ s1, s2 = line.split("|||")
979
+ yield line_id, {l1: s1.strip(), l2: s2.strip()}
980
+
981
+
982
+ def _parse_czeng(*paths, **kwargs):
983
+ """Generates examples from CzEng v1.6, with optional filtering for v1.7."""
984
+ filter_path = kwargs.get("filter_path", None)
985
+ if filter_path:
986
+ re_block = re.compile(r"^[^-]+-b(\d+)-\d\d[tde]")
987
+ with open(filter_path, encoding="utf-8") as f:
988
+ bad_blocks = {blk for blk in re.search(r"qw{([\s\d]*)}", f.read()).groups()[0].split()}
989
+ logging.info("Loaded %d bad blocks to filter from CzEng v1.6 to make v1.7.", len(bad_blocks))
990
+
991
+ for path in paths:
992
+ for gz_path in sorted(glob.glob(path)):
993
+ with open(gz_path, "rb") as g, gzip.GzipFile(fileobj=g) as f:
994
+ filename = os.path.basename(gz_path)
995
+ for line_id, line in enumerate(f):
996
+ line = line.decode("utf-8") # required for py3
997
+ if not line.strip():
998
+ continue
999
+ id_, unused_score, cs, en = line.split("\t")
1000
+ if filter_path:
1001
+ block_match = re.match(re_block, id_)
1002
+ if block_match and block_match.groups()[0] in bad_blocks:
1003
+ continue
1004
+ sub_key = "{}/{}".format(filename, line_id)
1005
+ yield sub_key, {
1006
+ "cs": cs.strip(),
1007
+ "en": en.strip(),
1008
+ }
1009
+
1010
+
1011
+ def _parse_hindencorp(path):
1012
+ with open(path, encoding="utf-8") as f:
1013
+ for line_id, line in enumerate(f):
1014
+ split_line = line.split("\t")
1015
+ if len(split_line) != 5:
1016
+ logging.warning("Skipping invalid HindEnCorp line: %s", line)
1017
+ continue
1018
+ yield line_id, {"translation": {"en": split_line[3].strip(), "hi": split_line[4].strip()}}