This commit is contained in:
Jakub Henyk 2024-04-22 21:57:49 +02:00
parent c9d855a803
commit 4807f6e442
7 changed files with 18209 additions and 0 deletions

8
.gitignore vendored Normal file
View File

@ -0,0 +1,8 @@
*~
*.swp
*.bak
*.pyc
*.o
.DS_Store
.token

10519
dev-0/out.tsv Normal file

File diff suppressed because it is too large Load Diff

7
generate_model.py Normal file
View File

@ -0,0 +1,7 @@
KENLM_BUILD_PATH='/home/ladislaus_iii/kenlm/build'
!$KENLM_BUILD_PATH/bin/lmplz -o 5 < train/in.txt > model.arpa
!$KENLM_BUILD_PATH/bin/build_binary model.arpa model.binary

79
predict.py Normal file
View File

@ -0,0 +1,79 @@
import kenlm
import csv
def predict_probability(sentence):
return model.score(sentence)
def load_candidate_words(file_path):
with open(file_path, 'r', encoding='utf-8') as file:
candidate_words = {line.strip() for line in file}
return candidate_words
def predict_word_between(text1, text2, model, candidate_words):
max_prob = float("-inf")
best_word = None
for word in candidate_words:
sentence = f"{text1} {word} {text2}"
prob = model.score(sentence)
if prob > max_prob:
max_prob = prob
best_word = word
return best_word
dev = []
test = []
with open('dev-0/in_1.csv', 'r', newline='', encoding='utf-8') as file:
reader = csv.reader(file, delimiter=',')
for row in reader:
dev.append(row)
with open('test-A/in_1.csv', 'r', newline='', encoding='utf-8') as file:
reader = csv.reader(file, delimiter=',')
for row in reader:
test.append(row)
model_path = "model.binary"
model = kenlm.Model(model_path)
candidate_words_file = "words_3.txt"
candidate_words = load_candidate_words(candidate_words_file)
predicted_dev = []
predicted_test = []
i = 0
for row in dev:
text1 = row[0]
text2 = row[1]
predicted_word = predict_word_between(text1, text2, model, candidate_words)
predicted_dev.append(predicted_word)
if i % 500 == 0:
print(f'{i/len(dev)*100}%')
i += 1
with open('dev-0/out.tsv', 'w', newline='') as tsv_file:
tsv_writer = csv.writer(tsv_file, delimiter='\t')
for row in predicted_dev:
tsv_writer.writerow(row)
i = 0
for row in test:
text1 = row[0]
text2 = row[1]
predicted_word = predict_word_between(text1, text2, model, candidate_words)
predicted_test.append(predicted_word)
if i % 500 == 0:
print(f'{i/len(dev)*100}%')
i += 1
with open('test-A/out.tsv', 'w', newline='') as tsv_file:
tsv_writer = csv.writer(tsv_file, delimiter='\t')
for row in predicted_test:
tsv_writer.writerow(row)

149
prep_data.py Normal file
View File

@ -0,0 +1,149 @@
import csv
import re
from gensim.models import Word2Vec
import gensim.downloader as api
import numpy as np
from spellchecker import SpellChecker
import pandas as pd
folder = 'test-A'
filename = f"{folder}/in_1.csv"
data = []
data = pd.read_csv(f'{folder}/in.tsv',delimiter='\t', header=None, encoding='utf-8', quoting=csv.QUOTE_NONE, engine='python').values.tolist()
data_a = []
data_b = []
data_pair = []
for i in range(len(data)):
data_a.append(data[i][6])
try:
data_b.append(data[i][7])
except:
data_b.append('')
for i in range(len(data)):
data_pair.append([data_a[i], data_b[i]])
data_tabs = []
for x, y in data_pair:
cleaned_text_a = x.replace('\\t', '\t').replace('\\n', '\n').strip("[]")
cleaned_text_b = y.replace('\\t', '\t').replace('\\n', '\n').strip("[]")
data_tabs.append([cleaned_text_a, cleaned_text_b])
data_removed = []
for x, y in data_tabs:
text = re.sub(r'(?<!-)\n', ' ', x)
text = re.sub(r'[\n-]', '', text)
text = re.sub(r'[^a-zA-Z0-9\s]', '', text)
text = re.sub(r'\s+', ' ', text)
text_2 = re.sub(r'(?<!-)\n', ' ', y)
text_2 = re.sub(r'[\n-]', '', text_2)
text_2 = re.sub(r'[^a-zA-Z0-9\s]', '', text_2)
text_2 = re.sub(r'\s+', ' ', text_2)
data_removed.append([text, text_2])
model = api.load("word2vec-google-news-300")
def is_close_to_actual(word, threshold=0.5):
if word in model:
similarities = model.similar_by_word(word)
return any(similarity > threshold for _, similarity in similarities)
else:
return False
def remove_words(text, words_to_destroy):
pattern = r'\b(?:{})\b'.format('|'.join(words_to_destroy))
cleaned_text = re.sub(pattern, '', text, flags=re.IGNORECASE)
return cleaned_text
spell = SpellChecker()
data_cleared = []
i = 0
for x, y in data_removed:
words = x.split()
words_2 = y.split()
misspelled = spell.unknown(words + words_2)
text = remove_words(x, list(misspelled))
text_2 = remove_words(y, list(misspelled))
data_cleared.append([text, text_2])
if i % 20000 == 0:
print(f'{i/430000*100}%')
i += 1
data_cleared_2 = []
for x, y in data_cleared:
text = re.sub(r'(?<!-)\n', ' ', x)
text = re.sub(r'[\n-]', '', text)
text = re.sub(r'[^a-zA-Z0-9\s]', '', text)
text = re.sub(r'\s+', ' ', text)
text_2 = re.sub(r'(?<!-)\n', ' ', y)
text_2 = re.sub(r'[\n-]', '', text_2)
text_2 = re.sub(r'[^a-zA-Z0-9\s]', '', text_2)
text_2 = re.sub(r'\s+', ' ', text_2)
data_cleared_2.append([text, text_2])
with open(filename, 'w', newline='') as csvfile:
writer = csv.writer(csvfile)
writer.writerows(data_cleared_2)
"""import wordninja
from spellchecker import SpellChecker
spell = SpellChecker()
concatenated_misspelled = []
for x, y in data_removed:
words = x.split()
words_2 = y.split()
misspelled = spell.unknown(words + words_2)
concatenated_misspelled.append(list(misspelled))
data_corrected = []
i = 0
for x, y in data_removed:
text = x
text_2 = y
for word in flattened_concatenated_misspelled:
if is_close_to_actual(word, model):
corrected_word = spell.correction(word)
if corrected_word != None:
text = text.replace(word, corrected_word)
text_2 = text_2.replace(word, corrected_word)
else:
if len(word) > 6:
tokens = wordninja.split(word)
my_string = ' '.join(tokens)
text = text.replace(word, my_string)
text_2 = text_2.replace(word, my_string)
else:
text = text.replace(word, '')
text_2 = text_2.replace(word, '')
if i % 20000 == 0:
print(f'{i/430000*100}%')
i += 1
data_corrected.append([text, text_2])"""

33
prep_txt.py Normal file
View File

@ -0,0 +1,33 @@
import csv
tr = []
tr_r = []
folder = 'dev-0'
with open(f'{folder}/in_1.csv', 'r', encoding='utf-8') as file:
csv_reader = csv.reader(file, delimiter=',')
for row in csv_reader:
tr.append(row)
with open(f'{folder}/expected.tsv', 'r', encoding='utf-8') as file:
csv_reader = csv.reader(file, delimiter='\t')
for row in csv_reader:
tr_r.append(row)
data = []
for i in range(len(tr)):
try:
data.append([tr[i][0], tr_r[i], tr[i][1]])
except:
try:
data.append([tr[i][0], tr_r[i], ''])
except:
pass
with open(f'{folder}/in.txt', 'w', encoding='utf-8') as f:
for item in data:
f.write(str(item[0]) + ' ' + str(item[1][0]) + ' ' + str(item[2]) + '\n')

7414
test-A/out.tsv Normal file

File diff suppressed because it is too large Load Diff