-
Notifications
You must be signed in to change notification settings - Fork 0
/
text-class-nn.py
183 lines (134 loc) · 6.29 KB
/
text-class-nn.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
178
179
180
181
182
183
import pandas as pd
import numpy as np
import string
import nltk
from keras_preprocessing.sequence import pad_sequences
from nltk.corpus import stopwords
from tensorflow.python.keras.layers import Conv1D, MaxPooling1D, Flatten
nltk.download('stopwords')
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout
from tensorflow.keras.layers import Embedding
from tensorflow.keras.layers import LSTM
from tensorflow.keras.preprocessing import sequence
from sklearn.model_selection import train_test_split
from keras.callbacks import CSVLogger
from keras.preprocessing.text import Tokenizer
def char_tokenizer(lines):
tk = Tokenizer(num_words=None, char_level=True, oov_token='UNK')
lines_low = [[l.lower() for l in lines]]
tk.fit_on_texts(lines_low)
alphabet = "abcdefghijklmnopqrstuvwxyz0123456789,;.!?:'\"/\\|_@#$%^&*~`+-=<>()[]{}"
char_dict = {}
for i, char in enumerate(alphabet):
char_dict[char] = i + 1
# Use char_dict to replace the tk.word_index
tk.word_index = char_dict.copy()
# Add 'UNK' to the vocabulary
tk.word_index[tk.oov_token] = max(char_dict.values()) + 1
return tk
def embed_weights(tk):
vocab_size = len(tk.word_index)
embedding_weights = []
embedding_weights.append(np.zeros(vocab_size))
for char,i in tk.word_index.items():
onehot = np.zeros(vocab_size)
onehot[i-1] = 1
embedding_weights.append(onehot)
return np.array(embedding_weights)
def unique_words(lines):
unique_words = set()
stop_words = set(stopwords.words('english'))
for line in lines:
for word in line:
if word not in stop_words:
unique_words.add(word)
word_vals = set(i for i in range(len(unique_words)))
word_keys = dict(zip(unique_words, word_vals))
return word_keys
def word_to_vec(posts):
stop_words = set(stopwords.words('english'))
posts = [line.lower().translate(str.maketrans('', '', string.punctuation)).split(' ') for line in posts]
word_dict = unique_words(posts)
word_key_map = [[word_dict[word] for word in post if word not in stop_words] for post in posts]
return word_key_map
def word_level(p_posts,n_posts):
neg_encoded = word_to_vec(n_posts)
pos_encoded = word_to_vec(p_posts)
posts = n_posts+p_posts
word_vecs = np.concatenate((neg_encoded, pos_encoded))
# Padding the data samples to a maximum review length in words
longest_post = max(len(post) for post in word_to_vec(posts))
padded_word_vecs = sequence.pad_sequences(word_vecs, maxlen=longest_post)
# re-create labels, concatenate with data, and set up test/train split
labels = np.concatenate((np.zeros(len(neg_encoded)), np.ones(len(pos_encoded))))
X_train, X_test, y_train, y_test = train_test_split(padded_word_vecs, labels, test_size=0.33)
print('X_train shape:', X_train.shape, y_train.shape)
print('X_test shape:', X_test.shape, y_test.shape)
##### LSTM
#todo:
# input dimension = Size of the vocabulary, i.e. maximum integer index + 1.
# adjust embedding size, number of epochs, plot val accuracy vs actual accuracy
# try other loss functions
vocab_size = max([max(arr) for arr in word_to_vec(posts)])
embedding_size = 128
model = Sequential()
model.add(Embedding(vocab_size, embedding_size, input_length=longest_post))
model.add(LSTM(128))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))
print(model.summary())
model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['accuracy'])
batch_size = 64
num_epochs = 10
X_valid, y_valid = X_train[:batch_size], y_train[:batch_size]
X_train2, y_train2 = X_train[batch_size:], y_train[batch_size:]
csv_logger = CSVLogger("model_history_log.csv", append=True)
model.fit(X_train2, y_train2, validation_data=(X_valid, y_valid), batch_size=batch_size, epochs=num_epochs,callbacks=[csv_logger])
scores = model.evaluate(X_test, y_test, verbose=0)
print('Test accuracy:', scores[1])
def char_level(p_posts,n_posts):
posts = p_posts + n_posts
tk = char_tokenizer(posts)
vocab_size = len(tk.word_index)
pos_seqs = tk.texts_to_sequences(p_posts)
neg_seqs = tk.texts_to_sequences(n_posts)
longest_post = max([len(x) for x in np.append(neg_seqs, pos_seqs, axis=0)])
pos_seq_padded = pad_sequences(pos_seqs, maxlen=longest_post, padding='post')
neg_seq_padded = pad_sequences(neg_seqs, maxlen=longest_post, padding='post')
pos_data = np.array(pos_seq_padded, dtype='float32')
neg_data = np.array(neg_seq_padded, dtype='float32')
padded_char_vecs = np.append(neg_data, pos_data, axis = 0)
labels = np.concatenate((np.zeros(len(neg_data)), np.ones(len(pos_data))))
X_train, X_test, y_train, y_test = train_test_split(padded_char_vecs, labels, test_size=0.33)
print('X_train shape:', X_train.shape, y_train.shape)
print('X_test shape:', X_test.shape, y_test.shape)
embedding_size = 69 #TODO: find where this number comes from
input_size = 1014
em_weights = embed_weights(tk)
embedding_layer = Embedding(vocab_size+1,embedding_size,input_length=input_size,weights=[em_weights])
cnn_model = Sequential()
cnn_model.add(embedding_layer)
cnn_model.add(Conv1D(embedding_size, kernel_size=8, activation='relu'))
cnn_model.add(MaxPooling1D(pool_size=2))
cnn_model.add(Flatten())
cnn_model.add(Dense(10, activation='relu'))
cnn_model.add(Dense(1, activation='sigmoid'))
cnn_model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
print(cnn_model.summary())
num_epochs = 5
cnn_model.fit(X_train, y_train, epochs=num_epochs, verbose=1)
scores = cnn_model.evaluate(X_test, y_test, verbose=0)
print('Test accuracy:', scores[1])
def main():
yelp_labelled = pd.read_csv('yelp_labelled.txt', sep='\t', header=None)
yelp_labelled.columns = ['text', 'label']
positive_posts = pd.Series.to_numpy(yelp_labelled[yelp_labelled.label == 1]['text'])
negative_posts = pd.Series.to_numpy(yelp_labelled[yelp_labelled.label == 0]['text'])
posts = positive_posts + negative_posts
##### WORD LEVEL NN #######
# word_level(positive_posts, negative_posts)
##### CHAR LEVEL NN ########
char_level(positive_posts, negative_posts)
if __name__ == "__main__":
main()