-
Notifications
You must be signed in to change notification settings - Fork 1
/
Siamese_Bidirectional_LSTM.py
180 lines (174 loc) · 8.14 KB
/
Siamese_Bidirectional_LSTM.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
# -*- coding: utf-8 -*-
"""
Created on Mon Sep 7 12:21:42 2020
@author: Rohit
"""
import matplotlib.pyplot as plt
import re
import nltk
import numpy as np
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer
from nltk.tokenize import word_tokenize
import seaborn as sns
from tqdm import tqdm
from sklearn.model_selection import train_test_split
import tensorflow as tf
import tensorflow.keras as keras
import tensorflow.keras.backend as K
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.layers import Input, Concatenate, Conv2D, Flatten, Dense, Embedding, LSTM, Bidirectional
from tensorflow.keras.models import Model
from tensorflow.keras.regularizers import l2
from tensorflow.keras.models import Sequential
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.layers import Conv2D, ZeroPadding2D, Activation, Input, concatenate
from tensorflow.keras.models import Model
import pandas as pd
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
from sklearn.preprocessing import MinMaxScaler, StandardScaler
pd.set_option('display.max_columns',20)
#preparing Dataset using title-paragraph.csv and citation data
title_paragraph = pd.read_csv('title_paragraph.csv')
col_names = ['hash', 'title_a', 'title_b', 'dist', 'count', 'page_idA', 'page_idB', 'cpi']
citation_pairs = pd.read_csv('citationdata5000000.csv', sep = '|',nrows = 1000000, header = None, names = col_names)
citation_pairs = pd.read_csv('processeddata.csv')
citation_pairs = citation_pairs[citation_pairs['dist'] <= 1000]
citation_pairs = citation_pairs.dropna(axis = 0)
text_a = pd.DataFrame(citation_pairs['title_a'])
text_a.rename(columns={'title_a':'title'}, inplace = True)
text_a_data = pd.merge(text_a, title_paragraph, how = 'left', on = 'title')
citation_pairs['text_a'] = text_a_data['text']
text_b = pd.DataFrame(citation_pairs['title_b'])
text_b.rename(columns={'title_b':'title'}, inplace = True)
text_b_data = pd.merge(text_b, title_paragraph, how = 'left', on = 'title')
citation_pairs['text_b'] = text_b_data['text']
citation_pairs = citation_pairs.dropna(axis = 0)
#Cleaning Data
def rephrase(phrase):
phrase = re.sub(r"won\'t", "will not", phrase)
phrase = re.sub(r"can\'t", "can not", phrase)
phrase = re.sub(r"n\'t", " not", phrase)
phrase = re.sub(r"\'re", " are", phrase)
phrase = re.sub(r"\'s", " is", phrase)
phrase = re.sub(r"\'d", " would", phrase)
phrase = re.sub(r"\'ll", " will", phrase)
phrase = re.sub(r"\'t", " not", phrase)
phrase = re.sub(r"\'ve", " have", phrase)
phrase = re.sub(r"\'m", " am", phrase)
return phrase
def stripunct(data):
return re.sub('[^A-Za-z]+', ' ', str(data), flags=re.MULTILINE|re.DOTALL)
stop_words = set(stopwords.words('english'))
stemm = WordNetLemmatizer()
def compute(sent):
sent = rephrase(sent)
sent = stripunct(sent)
words=word_tokenize(str(sent.lower()))
#Removing all single letter and and stopwords from question
sent1=' '.join(str(stemm.lemmatize(j)) for j in words if j not in stop_words and (len(j)!=1))
return sent1
clean_stemmed_text1 = []
clean_stemmed_text2 = []
combined_stemmed_text = []
for _, row in tqdm(citation_pairs.iterrows()):
csq1= compute(row['text_a'])
csq2= compute(row['text_b'])
clean_stemmed_text1.append(csq1)
clean_stemmed_text2.append(csq2)
combined_stemmed_text.append(csq1+" "+csq2)
citation_pairs['clean_stemmed_text1'] = clean_stemmed_text1
citation_pairs['clean_stemmed_text2'] = clean_stemmed_text2
citation_pairs['combined_stemmed_text'] = combined_stemmed_text
#Tokenization of text
token = Tokenizer()
token.fit_on_texts(citation_pairs['combined_stemmed_text'].values)
X_train, X_test, y_train, y_test = train_test_split(citation_pairs[['title_a','title_b','clean_stemmed_text1', 'clean_stemmed_text2']].tail(100000), citation_pairs['dist'].tail(100000), test_size=0.2, random_state=100)
X_train, X_val, y_train, y_val = train_test_split(X_train[['clean_stemmed_text1', 'clean_stemmed_text2']], y_train, test_size=0.2, random_state=100)
train_text1_seq = token.texts_to_sequences(X_train['clean_stemmed_text1'].values)
train_text2_seq = token.texts_to_sequences(X_train['clean_stemmed_text2'].values)
val_text1_seq = token.texts_to_sequences(X_val['clean_stemmed_text1'].values)
val_text2_seq = token.texts_to_sequences(X_val['clean_stemmed_text2'].values)
test_text1_seq = token.texts_to_sequences(X_test['clean_stemmed_text1'].values)
test_text2_seq = token.texts_to_sequences(X_test['clean_stemmed_text2'].values)
max_len = 50
train_text1_seq = pad_sequences(train_text1_seq, maxlen=max_len, padding='post')
train_text2_seq = pad_sequences(train_text2_seq, maxlen=max_len, padding='post')
val_text1_seq = pad_sequences(val_text1_seq, maxlen=max_len, padding='post')
val_text2_seq = pad_sequences(val_text2_seq, maxlen=max_len, padding='post')
test_text1_seq = pad_sequences(test_text1_seq, maxlen=max_len, padding='post')
test_text2_seq = pad_sequences(test_text2_seq, maxlen=max_len, padding='post')
#Creating Embedding Matrix
embeddings_index = {}
with open('glove.840B.300d.txt', encoding = 'UTF-8') as f:
for line in f:
values = line.split(' ')
word = values[0] # The word
coefficients = np.asarray(values[1:], dtype='float32')
embeddings_index[word] = coefficients
not_present_list = []
vocab_size = len(token.word_index) + 1
print('Loaded %s word vectors.' % len(embeddings_index))
embedding_matrix = np.zeros((vocab_size, len(embeddings_index['no'])))
for word, i in token.word_index.items():
if word in embeddings_index.keys():
embedding_vector = embeddings_index.get(word)
else:
not_present_list.append(word)
if embedding_vector is not None:
embedding_matrix[i] = embedding_vector
else:
embedding_matrix[i] = np.zeros((vocab_size, len(embeddings_index['no'])))
np.save('embedding_matrix.npy', embedding_matrix)
embedding_matrix = np.load('embedding_matrix.npy')
SC = StandardScaler()
y_train = SC.fit_transform(y_train.values.reshape(-1,1))
y_val = SC.fit_transform(y_val.values.reshape(-1,1))
y_test = SC.fit_transform(y_test.values.reshape(-1,1))
input_1 = Input(shape=(train_text1_seq.shape[1],))
input_2 = Input(shape=(train_text2_seq.shape[1],))
embed = Embedding(input_dim = vocab_size,
output_dim=300,weights=[embedding_matrix],
input_length=train_text1_seq.shape[1],trainable=False)
lstm_1 = embed(input_1)
lstm_2 = embed(input_2)
lstm = Bidirectional(LSTM(50, return_sequences = True, activation = 'relu'))
vector_1 = lstm(lstm_1)
vector_2 = lstm(lstm_2)
vector_1 = Flatten()(vector_1)
vector_2 = Flatten()(vector_2)
conc = concatenate([vector_1,vector_2])
out = Dense(1)(conc)
model = Model([input_1, input_2], out)
callback = [EarlyStopping(patience = 8)]
model.compile(optimizer=Adam(0.00001), loss='mse', metrics=['mae'])
model.load_weights('vcp8.h5')
history = model.fit([train_text1_seq,train_text2_seq],y_train.values.reshape(-1,1), epochs = 100,
batch_size=32,validation_data=([val_text1_seq, val_text2_seq],y_val.values.reshape(-1,1)),
callbacks = callback)
model.save_weights('vcp9.h5')
result = model.predict([test_text1_seq,test_text2_seq])
result = pd.DataFrame(result)
ytest = pd.DataFrame(y_test)
loss = history.history['val_loss']
epochs = range(1, len(loss) + 1)
mae_history_val = history.history['val_loss']
mae_history = history.history['loss']
plt.plot(epochs, mae_history,label='Training loss')
plt.plot(epochs, mae_history_val,label='Validation loss')
plt.title('Training and validation loss')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.show()
mae_history_val = history.history['val_mae']
mae_history = history.history['mae']
plt.plot(epochs, mae_history,label='Training mae')
plt.plot(epochs, mae_history_val,label='Validation mae')
plt.axhline(np.mean(mae_history), color = 'black', label = 'Mean MAE')
plt.title('Training and validation mae')
plt.xlabel('Epochs')
plt.ylabel('MAE')
plt.legend()
plt.show()