Add datasets

.gitattributes

@@ -25,3 +25,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zstandard filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.csv filter=lfs diff=lfs merge=lfs -text

clean_data.py

@@ -0,0 +1,58 @@
+import datasets
+from fastcore.utils import compose
+from clean_funcs import *
+
+fi_mc4 = datasets.load_dataset("mc4", "fi")
+print(fi_mc4)
+
+data_preprocessing_funcs = compose(*[fix_html, remove_control_char, remove_remaining_control_chars, remove_unicode_symbols,
+                                   standardise_punc, remove_news_tags, replace_urls, replace_usernames, remove_duplicate_words_punctuation, remove_multi_space])
+data_stats_funcs = compose(*[count_alphabet, count_numbers, count_upper, count_str_len,
+                           predict_lang, calculate_alphabet_ratio, calculate_number_ratio, calculate_upper_ratio])
+
+min_alphabet_ratio = 0.75
+max_upper_ratio = 0.10
+max_number_ratio = 0.05
+min_pred_lang_percentage = 0.95
+
+# TRAIN SPLIT
+num_rows = fi_mc4["train"].num_rows
+print(f"Original dataset train rows {num_rows}")
+fi_mc4["train"] = fi_mc4["train"].map(
+    data_preprocessing_funcs, num_proc=64, batched=True, writer_batch_size=100000)
+
+fi_train_only_longer = fi_mc4["train"].filter(
+    lambda example: len(example['text'].split()) >= 20, num_proc=64)
+num_rows = fi_train_only_longer.num_rows
+print(f"Only longer texts dataset train rows {num_rows}")
+
+fi_train_only_longer = fi_train_only_longer.map(
+    data_stats_funcs, num_proc=64, batched=False, writer_batch_size=100000)
+
+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[
+                                               'number_ratio'] < max_number_ratio and example['predicted_lang'] == '__label__fi' and example['predicted_lang_percentage'] > min_pred_lang_percentage, num_proc=64)
+num_rows = fi_train_cleaned.num_rows
+print(f"Final cleaned dataset train rows {num_rows}")
+
+# VAL SPLIT
+num_rows = fi_mc4["validation"].num_rows
+print(f"Original dataset val rows {num_rows}")
+fi_mc4["validation"] = fi_mc4["validation"].map(
+    data_preprocessing_funcs, num_proc=64, batched=True)
+
+fi_val_only_longer = fi_mc4["validation"].filter(
+    lambda example: len(example['text'].split()) >= 20, num_proc=64)
+num_rows = fi_val_only_longer.num_rows
+print(f"Only longer texts dataset val rows {num_rows}")
+
+fi_val_only_longer = fi_val_only_longer.map(
+    data_stats_funcs, num_proc=64, batched=False)
+
+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']
+                                           < max_number_ratio and example['predicted_lang'] == '__label__fi' and example['predicted_lang_percentage'] > min_pred_lang_percentage, num_proc=64)
+num_rows = fi_val_cleaned.num_rows
+print(f"Final cleaned dataset val rows {num_rows}")
+
+# SAVE TO DISK
+fi_train_cleaned.to_csv("train.csv", num_proc=64)
+fi_val_cleaned.to_csv("valid.csv", num_proc=64)

clean_funcs.py

@@ -0,0 +1,179 @@
+from fastcore.basics import listify
+import unicodedata
+import unidecode
+from string import punctuation
+import html
+from itertools import groupby
+import fasttext
+import re
+
+control_char_regex = re.compile(r'[\r\n\t]+')
+url_regex = re.compile(
+    r'((http|https)\:\/\/)?[a-zA-Z0-9\.\/\?\:@\-_=#]+\.([a-zA-Z]){2,6}([a-zA-Z0-9\.\&\/\?\:@\-_=#])*')
+username_regex = re.compile(r'(^|[^@\w])@(\w{1,15})\b')
+
+FASTTEXT_MODEL_PATH = 'lid.176.bin'
+fasttext_model = fasttext.load_model(FASTTEXT_MODEL_PATH)
+
+
+def fix_html(example):
+    "From fastai: 'Fix messy things we've seen in documents'"
+    tmp_ls = []
+    for e in listify(example['text']):
+        e = e.replace('#39;', "'").replace('amp;', '&').replace('#146;', "'").replace('nbsp;', ' ').replace(
+            '#36;', '$').replace('\\n', "\n").replace('quot;', "'").replace('<br />', "\n").replace(
+            '\\"', '"').replace('<unk>', ' ').replace(' @.@ ', '.').replace(' @-@ ', '-').replace('...', ' …')
+        tmp_ls.append(html.unescape(e))
+
+    example['text'] = tmp_ls
+    return example
+
+
+def remove_control_char(example):
+    tmp_ls = []
+    for e in listify(example['text']):
+        tmp_ls.append(re.sub(control_char_regex, '.', e))
+
+    example['text'] = tmp_ls
+    return example
+
+
+def remove_remaining_control_chars(example):
+    tmp_ls = []
+    for e in listify(example['text']):
+        tmp_ls.append(
+            ''.join(ch for ch in e if unicodedata.category(ch)[0] != 'C'))
+
+    example['text'] = tmp_ls
+    return example
+
+
+def remove_unicode_symbols(example):
+    tmp_ls = []
+    for e in listify(example['text']):
+        tmp_ls.append(
+            ''.join(ch for ch in e if unicodedata.category(ch)[0] != 'So'))
+
+    example['text'] = tmp_ls
+    return example
+
+
+def standardise_punc(example):
+    transl_table = dict([(ord(x), ord(y))
+                         for x, y in zip(u"‘’´“”–-",  u"'''\"\"--")])
+    tmp_ls = []
+    for e in listify(example['text']):
+        e = e.translate(transl_table)
+        e = re.sub(r"[^a-zA-Z0-9ÖÄÅöäå .,'%&€$=*@+;<>/()!?%:-]", " ", e)
+        tmp_ls.append(e)
+
+    example['text'] = tmp_ls
+    return example
+
+
+def remove_news_tags(example):
+    tmp_ls = []
+    for e in listify(example['text']):
+        e = re.sub(r"(<[A-Z].+?>)|(</[A-Z].+?>)", "", e)
+        tmp_ls.append(e)
+
+    example['text'] = tmp_ls
+    return example
+
+
+def replace_urls(example):
+    filler, tmp_ls = '', []
+    for e in listify(example['text']):
+        e = re.sub(r"(<a.+?>)|(</a>)|(<ref.+?>)", "", e)
+        e = re.sub(url_regex, filler, e)
+        tmp_ls.append(e)
+
+    example['text'] = tmp_ls
+    return example
+
+
+def replace_usernames(example):
+    filler, tmp_ls = '', []
+    for e in listify(example['text']):
+        occ = e.count('@')
+        for _ in range(occ):
+            e = e.replace('@<user>', f'{filler}')
+            # replace other user handles by filler
+            e = re.sub(username_regex, filler, e)
+            # add spaces between, and remove double spaces again
+            e = e.replace(filler, f' {filler} ')
+            e = ' '.join(e.split())
+        tmp_ls.append(e)
+
+    example['text'] = tmp_ls
+    return example
+
+
+def remove_duplicate_words_punctuation(example):
+    tmp_ls = []
+    for e in listify(example['text']):
+        e = re.sub(r'\b(\w+)( \1\b)+', r'\1', e)
+        punc = set(punctuation)
+        newtext = []
+        for k, g in groupby(e):
+            if k in punc:
+                newtext.append(k)
+            else:
+                newtext.extend(g)
+        e = ''.join(newtext)
+        tmp_ls.append(e)
+
+    example['text'] = tmp_ls
+    return example
+
+
+def remove_multi_space(example):
+    tmp_ls = []
+    for e in listify(example['text']):
+        tmp_ls.append(' '.join(e.split()))
+
+    example['text'] = tmp_ls
+    return example
+
+
+def count_alphabet(batch):
+    batch['alphabet_len'] = len(re.findall(r'[äÄöÖåÅa-zA-Z]', batch['text']))
+    return batch
+
+
+def count_numbers(batch):
+    batch['number_len'] = len(re.findall(r'[0-9]', batch['text']))
+    return batch
+
+
+def count_upper(batch):
+    batch['upper_len'] = len(re.findall(r'[ÄÖÅA-Z]', batch['text']))
+    return batch
+
+
+def count_str_len(batch):
+    batch['total_len'] = len(batch['text'])
+    return batch
+
+
+def predict_lang(batch):
+    pred = fasttext_model.predict(batch['text'])
+    batch['predicted_lang'] = pred[0][0]
+    batch['predicted_lang_percentage'] = float(pred[1][0])
+    return batch
+
+
+def calculate_alphabet_ratio(batch):
+    batch['alphabet_ratio'] = int(
+        batch['alphabet_len']) / int(batch['total_len'])
+    return batch
+
+
+def calculate_number_ratio(batch):
+    batch['number_ratio'] = int(batch['number_len']) / int(batch['total_len'])
+    return batch
+
+
+def calculate_upper_ratio(batch):
+    batch['upper_ratio'] = int(batch['upper_len']) / int(batch['total_len'])
+    return batch

train.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc9f4e2d48ba5fdd16e8636c17f775bcab1284958705b7cbb097005d1fa8f579
+size 66554165333

valid.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff43367cf849f61f749cad72307c8bbdb67b46f553c004f35ad07ca683a83a9a
+size 65904851