-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathlanguage_model.py
177 lines (146 loc) · 7.17 KB
/
language_model.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
import nltk
import math
import random
from utils.smoothing import smoothing, count_occurrences
from collections import Counter
from utils.process import END_OF_SENTENCE, END_OF_WORD
class LanguageModel():
def __init__(self, tokens):
self._init_unigram(tokens)
self._init_bigram()
self._init_trigram()
def _init_unigram(self, tokens):
print("Setting unigrams...")
self.tokens = tokens
self.unigrams = list((nltk.ngrams(self.tokens, 1)))
self.unigram_table = Counter(self.unigrams)
self.unigram_count_occurrences = count_occurrences(self.unigram_table)
print("Applying smoothing to unigrams...")
self.unigram_frequencies, self.unigram_zero_counts = smoothing(self.unigram_count_occurrences, self.unigram_table)
with open('output/unigram_frequencies', 'w', encoding='utf-8') as file:
for key, value in self.unigram_frequencies.items():
key_str = ', '.join(key)
file.write(f'{key_str}: {value}\n')
def _init_bigram(self):
print("Setting bigrams...")
self.bigrams = list((nltk.ngrams(self.tokens, 2)))
self.bigram_table = Counter(self.bigrams)
self.bigram_count_occurrences = count_occurrences(self.bigram_table)
print("Applying smoothing to bigrams...")
self.bigram_frequencies, self.bigram_zero_counts = smoothing(self.bigram_count_occurrences, self.bigram_table)
with open('output/bigram_frequencies', 'w', encoding='utf-8') as file:
for key, value in self.bigram_frequencies.items():
key_str = ', '.join(key)
file.write(f'{key_str}: {value}\n')
def _init_trigram(self):
print("Setting trigrams...")
self.trigrams = list((nltk.ngrams(self.tokens, 3)))
self.trigram_table = Counter(self.trigrams)
self.trigram_count_occurrences = count_occurrences(self.trigram_table)
print("Applying smoothing to trigrams...")
self.trigram_frequencies, self.trigram_zero_counts = smoothing(self.trigram_count_occurrences, self.trigram_table)
with open('output/trigram_frequencies', 'w', encoding='utf-8') as file:
for key, value in self.trigram_frequencies.items():
key_str = ', '.join(key)
file.write(f'{key_str}: {value}\n')
def _unigram_probability(self, syllable):
numerator = self.unigram_frequencies.get(syllable, self.unigram_zero_counts)
denominator = len(self.unigrams)
return float(numerator) / float(denominator)
def _bigram_probability(self, prev, curr):
numerator = self.bigram_frequencies.get((prev, curr), self.bigram_zero_counts)
denominator = self.unigram_frequencies.get((prev,), self.unigram_zero_counts)
return float(numerator) / float(denominator)
def _trigram_probability(self, prev1, prev2, curr):
numerator = self.trigram_frequencies.get((prev1, prev2, curr), self.trigram_zero_counts)
denominator = self.bigram_frequencies.get((prev1, prev2), self.bigram_zero_counts)
return float(numerator) / float(denominator)
def unigram_perplexity(self, tokens):
number_of_tokens = len(tokens)
if number_of_tokens < 1:
return 0
prob_sum_logs = 0
for syllable in tokens:
syllable_probability = self._unigram_probability(syllable)
prob_sum_logs += math.log(syllable_probability)
perplexity = math.exp(-(prob_sum_logs / number_of_tokens))
return perplexity
def bigram_perplexity(self, tokens):
number_of_tokens = len(tokens)
if number_of_tokens < 2:
return 0
prob_sum_logs = 0
prev = None
for curr in tokens:
if prev is not None:
syllable_probability = self._bigram_probability(prev, curr)
prob_sum_logs += math.log(syllable_probability)
prev = curr
perplexity = math.exp(-(prob_sum_logs / (number_of_tokens - 1)))
return perplexity
def trigram_perplexity(self, tokens):
number_of_tokens = len(tokens)
if number_of_tokens < 3:
return 0
prob_sum_logs = 0
prev1, prev2 = None, None
for curr in tokens:
if prev1 is not None and prev2 is not None:
syllable_probability = self._trigram_probability(prev1, prev2, curr)
prob_sum_logs += math.log(syllable_probability)
prev1, prev2 = prev2, curr
perplexity = math.exp(-(prob_sum_logs / (number_of_tokens - 2)))
return perplexity
def generate_random_sentence_unigram(self, n=5):
sentence = []
start_word = END_OF_SENTENCE
sentence.append(start_word)
total_frequency = sum(value for key, value in self.unigram_frequencies.items() if key[0] not in [END_OF_SENTENCE, END_OF_WORD])
words = [key[0] for key in self.unigram_frequencies.keys() if key[0] not in [END_OF_SENTENCE, END_OF_WORD]]
weights = [value / total_frequency for key, value in self.unigram_frequencies.items() if key[0] not in [END_OF_SENTENCE, END_OF_WORD]]
for _ in range(n):
next_word = random.choices(
words,
weights=weights,
k=1
)[0]
sentence.append(next_word)
sentence.append(END_OF_SENTENCE)
return ' '.join(sentence)
def generate_random_sentence_bigram(self, n=5):
sentence = []
start_word = END_OF_SENTENCE
sentence.append(start_word)
for _ in range(n):
next_word_choices = [key[1] for key in self.bigram_frequencies.keys() if key[0] == start_word]
weights = [self.bigram_frequencies[key] for key in self.bigram_frequencies.keys() if key[0] == start_word]
if not next_word_choices:
break
next_word = random.choices(
next_word_choices,
weights=weights,
k=1
)[0]
sentence.append(next_word)
start_word = next_word
sentence.append(END_OF_SENTENCE)
return ' '.join(sentence)
def generate_random_sentence_trigram(self, n=5):
sentence = []
start_words = random.choice([key for key in self.trigram_frequencies.keys() if key[0] == END_OF_SENTENCE and key[1] != END_OF_SENTENCE])
sentence.extend(start_words)
for _ in range(n - 2):
prev_words = tuple(sentence[-2:])
next_word_choices = [word[2] for word in self.trigram_frequencies.keys() if word[:2] == prev_words and word[2] != END_OF_SENTENCE]
weights = [self.trigram_frequencies[word] for word in self.trigram_frequencies.keys() if word[:2] == prev_words and word[2] != END_OF_SENTENCE]
if next_word_choices:
next_word = random.choices(
next_word_choices,
weights=weights,
k=1
)[0]
sentence.append(next_word)
else:
break
sentence.append(END_OF_SENTENCE)
return ' '.join(sentence)