Datasets:
aapot
commited on
Commit
•
05759f3
1
Parent(s):
0205cf8
Add datasets
Browse files- .gitattributes +1 -0
- clean_data.py +58 -0
- clean_funcs.py +179 -0
- train.csv +3 -0
- valid.csv +3 -0
.gitattributes
CHANGED
@@ -25,3 +25,4 @@ saved_model/**/* 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
|
|
|
|
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
|
28 |
+
*.csv filter=lfs diff=lfs merge=lfs -text
|
clean_data.py
ADDED
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import datasets
|
2 |
+
from fastcore.utils import compose
|
3 |
+
from clean_funcs import *
|
4 |
+
|
5 |
+
fi_mc4 = datasets.load_dataset("mc4", "fi")
|
6 |
+
print(fi_mc4)
|
7 |
+
|
8 |
+
data_preprocessing_funcs = compose(*[fix_html, remove_control_char, remove_remaining_control_chars, remove_unicode_symbols,
|
9 |
+
standardise_punc, remove_news_tags, replace_urls, replace_usernames, remove_duplicate_words_punctuation, remove_multi_space])
|
10 |
+
data_stats_funcs = compose(*[count_alphabet, count_numbers, count_upper, count_str_len,
|
11 |
+
predict_lang, calculate_alphabet_ratio, calculate_number_ratio, calculate_upper_ratio])
|
12 |
+
|
13 |
+
min_alphabet_ratio = 0.75
|
14 |
+
max_upper_ratio = 0.10
|
15 |
+
max_number_ratio = 0.05
|
16 |
+
min_pred_lang_percentage = 0.95
|
17 |
+
|
18 |
+
# TRAIN SPLIT
|
19 |
+
num_rows = fi_mc4["train"].num_rows
|
20 |
+
print(f"Original dataset train rows {num_rows}")
|
21 |
+
fi_mc4["train"] = fi_mc4["train"].map(
|
22 |
+
data_preprocessing_funcs, num_proc=64, batched=True, writer_batch_size=100000)
|
23 |
+
|
24 |
+
fi_train_only_longer = fi_mc4["train"].filter(
|
25 |
+
lambda example: len(example['text'].split()) >= 20, num_proc=64)
|
26 |
+
num_rows = fi_train_only_longer.num_rows
|
27 |
+
print(f"Only longer texts dataset train rows {num_rows}")
|
28 |
+
|
29 |
+
fi_train_only_longer = fi_train_only_longer.map(
|
30 |
+
data_stats_funcs, num_proc=64, batched=False, writer_batch_size=100000)
|
31 |
+
|
32 |
+
fi_train_cleaned = fi_train_only_longer.filter(lambda example: example['alphabet_ratio'] > min_alphabet_ratio and example['upper_ratio'] < max_upper_ratio and example[
|
33 |
+
'number_ratio'] < max_number_ratio and example['predicted_lang'] == '__label__fi' and example['predicted_lang_percentage'] > min_pred_lang_percentage, num_proc=64)
|
34 |
+
num_rows = fi_train_cleaned.num_rows
|
35 |
+
print(f"Final cleaned dataset train rows {num_rows}")
|
36 |
+
|
37 |
+
# VAL SPLIT
|
38 |
+
num_rows = fi_mc4["validation"].num_rows
|
39 |
+
print(f"Original dataset val rows {num_rows}")
|
40 |
+
fi_mc4["validation"] = fi_mc4["validation"].map(
|
41 |
+
data_preprocessing_funcs, num_proc=64, batched=True)
|
42 |
+
|
43 |
+
fi_val_only_longer = fi_mc4["validation"].filter(
|
44 |
+
lambda example: len(example['text'].split()) >= 20, num_proc=64)
|
45 |
+
num_rows = fi_val_only_longer.num_rows
|
46 |
+
print(f"Only longer texts dataset val rows {num_rows}")
|
47 |
+
|
48 |
+
fi_val_only_longer = fi_val_only_longer.map(
|
49 |
+
data_stats_funcs, num_proc=64, batched=False)
|
50 |
+
|
51 |
+
fi_val_cleaned = fi_val_only_longer.filter(lambda example: example['alphabet_ratio'] > min_alphabet_ratio and example['upper_ratio'] < max_upper_ratio and example['number_ratio']
|
52 |
+
< max_number_ratio and example['predicted_lang'] == '__label__fi' and example['predicted_lang_percentage'] > min_pred_lang_percentage, num_proc=64)
|
53 |
+
num_rows = fi_val_cleaned.num_rows
|
54 |
+
print(f"Final cleaned dataset val rows {num_rows}")
|
55 |
+
|
56 |
+
# SAVE TO DISK
|
57 |
+
fi_train_cleaned.to_csv("train.csv", num_proc=64)
|
58 |
+
fi_val_cleaned.to_csv("valid.csv", num_proc=64)
|
clean_funcs.py
ADDED
@@ -0,0 +1,179 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from fastcore.basics import listify
|
2 |
+
import unicodedata
|
3 |
+
import unidecode
|
4 |
+
from string import punctuation
|
5 |
+
import html
|
6 |
+
from itertools import groupby
|
7 |
+
import fasttext
|
8 |
+
import re
|
9 |
+
|
10 |
+
control_char_regex = re.compile(r'[\r\n\t]+')
|
11 |
+
url_regex = re.compile(
|
12 |
+
r'((http|https)\:\/\/)?[a-zA-Z0-9\.\/\?\:@\-_=#]+\.([a-zA-Z]){2,6}([a-zA-Z0-9\.\&\/\?\:@\-_=#])*')
|
13 |
+
username_regex = re.compile(r'(^|[^@\w])@(\w{1,15})\b')
|
14 |
+
|
15 |
+
FASTTEXT_MODEL_PATH = 'lid.176.bin'
|
16 |
+
fasttext_model = fasttext.load_model(FASTTEXT_MODEL_PATH)
|
17 |
+
|
18 |
+
|
19 |
+
def fix_html(example):
|
20 |
+
"From fastai: 'Fix messy things we've seen in documents'"
|
21 |
+
tmp_ls = []
|
22 |
+
for e in listify(example['text']):
|
23 |
+
e = e.replace('#39;', "'").replace('amp;', '&').replace('#146;', "'").replace('nbsp;', ' ').replace(
|
24 |
+
'#36;', '$').replace('\\n', "\n").replace('quot;', "'").replace('<br />', "\n").replace(
|
25 |
+
'\\"', '"').replace('<unk>', ' ').replace(' @.@ ', '.').replace(' @-@ ', '-').replace('...', ' …')
|
26 |
+
tmp_ls.append(html.unescape(e))
|
27 |
+
|
28 |
+
example['text'] = tmp_ls
|
29 |
+
return example
|
30 |
+
|
31 |
+
|
32 |
+
def remove_control_char(example):
|
33 |
+
tmp_ls = []
|
34 |
+
for e in listify(example['text']):
|
35 |
+
tmp_ls.append(re.sub(control_char_regex, '.', e))
|
36 |
+
|
37 |
+
example['text'] = tmp_ls
|
38 |
+
return example
|
39 |
+
|
40 |
+
|
41 |
+
def remove_remaining_control_chars(example):
|
42 |
+
tmp_ls = []
|
43 |
+
for e in listify(example['text']):
|
44 |
+
tmp_ls.append(
|
45 |
+
''.join(ch for ch in e if unicodedata.category(ch)[0] != 'C'))
|
46 |
+
|
47 |
+
example['text'] = tmp_ls
|
48 |
+
return example
|
49 |
+
|
50 |
+
|
51 |
+
def remove_unicode_symbols(example):
|
52 |
+
tmp_ls = []
|
53 |
+
for e in listify(example['text']):
|
54 |
+
tmp_ls.append(
|
55 |
+
''.join(ch for ch in e if unicodedata.category(ch)[0] != 'So'))
|
56 |
+
|
57 |
+
example['text'] = tmp_ls
|
58 |
+
return example
|
59 |
+
|
60 |
+
|
61 |
+
def standardise_punc(example):
|
62 |
+
transl_table = dict([(ord(x), ord(y))
|
63 |
+
for x, y in zip(u"‘’´“”–-", u"'''\"\"--")])
|
64 |
+
tmp_ls = []
|
65 |
+
for e in listify(example['text']):
|
66 |
+
e = e.translate(transl_table)
|
67 |
+
e = re.sub(r"[^a-zA-Z0-9ÖÄÅöäå .,'%&€$=*@+;<>/()!?%:-]", " ", e)
|
68 |
+
tmp_ls.append(e)
|
69 |
+
|
70 |
+
example['text'] = tmp_ls
|
71 |
+
return example
|
72 |
+
|
73 |
+
|
74 |
+
def remove_news_tags(example):
|
75 |
+
tmp_ls = []
|
76 |
+
for e in listify(example['text']):
|
77 |
+
e = re.sub(r"(<[A-Z].+?>)|(</[A-Z].+?>)", "", e)
|
78 |
+
tmp_ls.append(e)
|
79 |
+
|
80 |
+
example['text'] = tmp_ls
|
81 |
+
return example
|
82 |
+
|
83 |
+
|
84 |
+
def replace_urls(example):
|
85 |
+
filler, tmp_ls = '', []
|
86 |
+
for e in listify(example['text']):
|
87 |
+
e = re.sub(r"(<a.+?>)|(</a>)|(<ref.+?>)", "", e)
|
88 |
+
e = re.sub(url_regex, filler, e)
|
89 |
+
tmp_ls.append(e)
|
90 |
+
|
91 |
+
example['text'] = tmp_ls
|
92 |
+
return example
|
93 |
+
|
94 |
+
|
95 |
+
def replace_usernames(example):
|
96 |
+
filler, tmp_ls = '', []
|
97 |
+
for e in listify(example['text']):
|
98 |
+
occ = e.count('@')
|
99 |
+
for _ in range(occ):
|
100 |
+
e = e.replace('@<user>', f'{filler}')
|
101 |
+
# replace other user handles by filler
|
102 |
+
e = re.sub(username_regex, filler, e)
|
103 |
+
# add spaces between, and remove double spaces again
|
104 |
+
e = e.replace(filler, f' {filler} ')
|
105 |
+
e = ' '.join(e.split())
|
106 |
+
tmp_ls.append(e)
|
107 |
+
|
108 |
+
example['text'] = tmp_ls
|
109 |
+
return example
|
110 |
+
|
111 |
+
|
112 |
+
def remove_duplicate_words_punctuation(example):
|
113 |
+
tmp_ls = []
|
114 |
+
for e in listify(example['text']):
|
115 |
+
e = re.sub(r'\b(\w+)( \1\b)+', r'\1', e)
|
116 |
+
punc = set(punctuation)
|
117 |
+
newtext = []
|
118 |
+
for k, g in groupby(e):
|
119 |
+
if k in punc:
|
120 |
+
newtext.append(k)
|
121 |
+
else:
|
122 |
+
newtext.extend(g)
|
123 |
+
e = ''.join(newtext)
|
124 |
+
tmp_ls.append(e)
|
125 |
+
|
126 |
+
example['text'] = tmp_ls
|
127 |
+
return example
|
128 |
+
|
129 |
+
|
130 |
+
def remove_multi_space(example):
|
131 |
+
tmp_ls = []
|
132 |
+
for e in listify(example['text']):
|
133 |
+
tmp_ls.append(' '.join(e.split()))
|
134 |
+
|
135 |
+
example['text'] = tmp_ls
|
136 |
+
return example
|
137 |
+
|
138 |
+
|
139 |
+
def count_alphabet(batch):
|
140 |
+
batch['alphabet_len'] = len(re.findall(r'[äÄöÖåÅa-zA-Z]', batch['text']))
|
141 |
+
return batch
|
142 |
+
|
143 |
+
|
144 |
+
def count_numbers(batch):
|
145 |
+
batch['number_len'] = len(re.findall(r'[0-9]', batch['text']))
|
146 |
+
return batch
|
147 |
+
|
148 |
+
|
149 |
+
def count_upper(batch):
|
150 |
+
batch['upper_len'] = len(re.findall(r'[ÄÖÅA-Z]', batch['text']))
|
151 |
+
return batch
|
152 |
+
|
153 |
+
|
154 |
+
def count_str_len(batch):
|
155 |
+
batch['total_len'] = len(batch['text'])
|
156 |
+
return batch
|
157 |
+
|
158 |
+
|
159 |
+
def predict_lang(batch):
|
160 |
+
pred = fasttext_model.predict(batch['text'])
|
161 |
+
batch['predicted_lang'] = pred[0][0]
|
162 |
+
batch['predicted_lang_percentage'] = float(pred[1][0])
|
163 |
+
return batch
|
164 |
+
|
165 |
+
|
166 |
+
def calculate_alphabet_ratio(batch):
|
167 |
+
batch['alphabet_ratio'] = int(
|
168 |
+
batch['alphabet_len']) / int(batch['total_len'])
|
169 |
+
return batch
|
170 |
+
|
171 |
+
|
172 |
+
def calculate_number_ratio(batch):
|
173 |
+
batch['number_ratio'] = int(batch['number_len']) / int(batch['total_len'])
|
174 |
+
return batch
|
175 |
+
|
176 |
+
|
177 |
+
def calculate_upper_ratio(batch):
|
178 |
+
batch['upper_ratio'] = int(batch['upper_len']) / int(batch['total_len'])
|
179 |
+
return batch
|
train.csv
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:fc9f4e2d48ba5fdd16e8636c17f775bcab1284958705b7cbb097005d1fa8f579
|
3 |
+
size 66554165333
|
valid.csv
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ff43367cf849f61f749cad72307c8bbdb67b46f553c004f35ad07ca683a83a9a
|
3 |
+
size 65904851
|