joelniklaus commited on
Commit
30dd125
1 Parent(s): 3ffe3b2

added data loader script and data preparation script

Browse files
Files changed (2) hide show
  1. EU_Wikipedias.py +134 -0
  2. prepare_wikipedias.py +48 -0
EU_Wikipedias.py ADDED
@@ -0,0 +1,134 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """EU Wikipedias"""
2
+
3
+ import json
4
+
5
+ import datasets
6
+ from huggingface_hub.file_download import hf_hub_url
7
+
8
+ try:
9
+ import lzma as xz
10
+ except ImportError:
11
+ import pylzma as xz
12
+
13
+ datasets.logging.set_verbosity_info()
14
+ logger = datasets.logging.get_logger(__name__)
15
+
16
+ _CITATION = """\
17
+ @ONLINE {wikidump,
18
+ author = {Wikimedia Foundation},
19
+ title = {Wikimedia Downloads},
20
+ url = {https://dumps.wikimedia.org}
21
+ }
22
+ """
23
+
24
+ _DESCRIPTION = """\
25
+ Wikipedia dataset containing cleaned articles of all languages.
26
+ The datasets are built from the Wikipedia dump
27
+ (https://dumps.wikimedia.org/) with one split per language. Each example
28
+ contains the content of one full Wikipedia article with cleaning to strip
29
+ markdown and unwanted sections (references, etc.).
30
+ """
31
+
32
+ _LICENSE = (
33
+ "This work is licensed under the Creative Commons Attribution-ShareAlike "
34
+ "3.0 Unported License. To view a copy of this license, visit "
35
+ "http://creativecommons.org/licenses/by-sa/3.0/ or send a letter to "
36
+ "Creative Commons, PO Box 1866, Mountain View, CA 94042, USA."
37
+ )
38
+
39
+ _URL = "https://huggingface.co/datasets/joelito/EU_Wikipedias"
40
+
41
+ _LANGUAGES = ["bg", "cs", "da", "de", "el", "en", "es", "et", "fi", "fr", "ga", "hr",
42
+ "hu", "it", "lt", "lv", "mt", "nl", "pl", "pt", "ro", "sk", "sl", "sv"]
43
+
44
+ _DATES = ["20221120"] # one can add more in the future with the file prepare_wikipedias.py
45
+
46
+
47
+ class EUWikipediasConfig(datasets.BuilderConfig):
48
+ """BuilderConfig for EUWikipedias."""
49
+
50
+ def __init__(self, date=None, language=None, **kwargs):
51
+ """BuilderConfig for EUWikipedias.
52
+ Args:
53
+ language: string, the language code for the Wikipedia dump to use:
54
+ One of bg,cs,da,de,el,en,es,et,fi,fr,ga,hr,hu,it,lt,lv,mt,nl,pl,pt,ro,sk,sl,sv or all
55
+ date: string, date of the Wikipedia dump in YYYYMMDD format. A list of
56
+ available dates can be found at https://dumps.wikimedia.org/enwiki/.
57
+ **kwargs: keyword arguments forwarded to super.
58
+ """
59
+ if date not in _DATES:
60
+ raise ValueError(f"date must be one of {_DATES} but was `{date}`")
61
+ if language not in _LANGUAGES + ["all"]:
62
+ raise ValueError(f"language must be one of {_LANGUAGES} but was `{language}`")
63
+
64
+ super().__init__(
65
+ name=f"{date}.{language}",
66
+ description=f"Wikipedia dataset for {language}, parsed from {date} dump.",
67
+ **kwargs,
68
+ )
69
+ self.date = date
70
+ self.language = language
71
+
72
+
73
+ class EUWikipedias(datasets.GeneratorBasedBuilder):
74
+ """EUWikipedias: A dataset of Wikipedias in the EU languages"""
75
+ BUILDER_CONFIG_CLASS = EUWikipediasConfig
76
+
77
+ BUILDER_CONFIGS = [EUWikipediasConfig(date=date, language=language)
78
+ for language in _LANGUAGES + ["all"]
79
+ for date in _DATES]
80
+
81
+ def _info(self):
82
+ return datasets.DatasetInfo(
83
+ description=_DESCRIPTION,
84
+ features=datasets.Features(
85
+ {
86
+ "language": datasets.Value("string"),
87
+ "id": datasets.Value("string"),
88
+ "url": datasets.Value("string"),
89
+ "title": datasets.Value("string"),
90
+ "text": datasets.Value("string"),
91
+ }
92
+ ),
93
+ supervised_keys=None, # No default supervised_keys.
94
+ homepage=_URL,
95
+ citation=_CITATION,
96
+ )
97
+
98
+ def _split_generators(self, dl_manager):
99
+ def download_url(dataset, file_name):
100
+ url = hf_hub_url(repo_id=dataset, filename=f"data/{file_name}.jsonl.xz", repo_type="dataset")
101
+ return dl_manager.download(url)
102
+
103
+ data_infos = []
104
+ languages = _LANGUAGES if self.config.language == "all" else [self.config.language]
105
+
106
+ for language in languages:
107
+ info = {"language": language}
108
+ for shard in range(50): # 50 shards per language at most (English has 10 in 20221120)
109
+ try:
110
+ info["filepath"] = download_url("joelito/EU_Wikipedias", f"{self.config.date}/{language}_{shard}")
111
+ data_infos.append(info.copy())
112
+ except:
113
+ break # we found the last shard
114
+
115
+ return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"data_infos": data_infos})]
116
+
117
+ def _generate_examples(self, data_infos):
118
+ """This function returns the examples in the raw (text) form by iterating on all the files."""
119
+ id_ = 0
120
+ for data_info in data_infos:
121
+ logger.info("Generating examples from = %s", data_info["filepath"])
122
+ try:
123
+ with xz.open(open(data_info["filepath"], "rb"), "rt", encoding="utf-8") as f:
124
+ for line in f:
125
+ if line:
126
+ example = json.loads(line)
127
+ if example is not None and isinstance(example, dict):
128
+ yield id_, {
129
+ "language": data_info["language"], # add the language
130
+ **example,
131
+ }
132
+ id_ += 1
133
+ except Exception:
134
+ logger.exception("Error while processing file %s", data_info["filepath"])
prepare_wikipedias.py ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import glob
2
+ import os
3
+
4
+ from datasets import load_dataset
5
+
6
+ _LANGUAGES = ['bg', 'cs', 'da', 'de', 'el', 'en', 'es', 'et', 'fi', 'fr', 'ga', 'hr',
7
+ 'hu', 'it', 'lt', 'lv', 'mt', 'nl', 'pl', 'pt', 'ro', 'sk', 'sl', 'sv']
8
+ ERROR_LANGS = ['et', 'ga'] # 20221101: ga somehow hangs at sample 36715/57711, for et no data is downloaded
9
+ ERROR_LANGS = [] # 20221120: no errors here
10
+ FINAL_LANGS = [l for l in _LANGUAGES if l not in ERROR_LANGS]
11
+ date = "20221120" # every 01 and 20 of the month, just edit this to generate newer data
12
+
13
+ base_dir = 'data'
14
+ date_dir = os.path.join(base_dir, date)
15
+ os.makedirs(date_dir, exist_ok=True)
16
+
17
+ max_file_size = 4 # GB
18
+
19
+
20
+ def process_language(LANG):
21
+ print(f'Processing language {LANG}...')
22
+ # streaming does not work here!
23
+ dataset = load_dataset("olm/wikipedia", language=LANG, date=date, split='train')
24
+ size_in_gb = dataset.size_in_bytes / 1e9
25
+ print(f'Found {size_in_gb} GB of data ({len(dataset)} documents) for language {LANG}...')
26
+ if size_in_gb > max_file_size:
27
+ num_shards = int(size_in_gb / max_file_size) + 1
28
+ for shard in range(num_shards):
29
+ dataset.shard(num_shards, shard).to_json(f'{date_dir}/{LANG}_{shard}.jsonl', lines=True)
30
+ else:
31
+ dataset.to_json(f'{date_dir}/{LANG}_0.jsonl', lines=True)
32
+
33
+
34
+ if __name__ == '__main__':
35
+ """
36
+ Run with
37
+ export PYTHONPATH=. && python prepare_wikipedias.py | tee prepare_wikipedias.log
38
+ """
39
+ # it does not work in parallel
40
+ for LANG in FINAL_LANGS:
41
+ process_language(LANG)
42
+
43
+ # Compress datasets
44
+ print(f"Compressing datasets at {date_dir}")
45
+ # Do this at the end because we use multithreading
46
+ for path in glob.glob(os.path.join(date_dir, '*.jsonl')):
47
+ os.system(f'xz -zkf -T0 {path}') # -TO to use multithreading
48
+ # os.system(f'rm {path}') # remove uncompressed file to save space