-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathclassifier_models.py
204 lines (179 loc) · 10.2 KB
/
classifier_models.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
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
import numpy as np
import pandas as pd
from keras.preprocessing.text import Tokenizer
from keras.preprocessing.sequence import pad_sequences
from tensorflow.keras import regularizers
import tensorflow as tf
import tensorflow_hub as hub
print("tensorflow version : ", tf.__version__)
print("tensorflow_hub version : ", hub.__version__)
from sklearn.metrics import classification_report, accuracy_score, precision_score, recall_score, f1_score, cohen_kappa_score, roc_auc_score, confusion_matrix
from sklearn.utils import resample
MAX_NB_WORDS = 50000
vocabulary_size=MAX_NB_WORDS
EMBEDDING_DIM = 300
TOPICS_LEN = 14
TEXT_LEN = 1000
ENTITIES_LEN = 1000
TRIPLES_LEN = 1000
def model_making(count, embedding_matrix, sents=False, topics=False, entities=False, triples=False, text=False):
learning_rate = 2e-5
mod_out=[]
mod_in=[]
dropout_rate = 0.5
if (text==True):
input_text = tf.keras.layers.Input(shape=(TEXT_LEN,), name='input_text')
m1_layers = tf.keras.layers.Embedding(MAX_NB_WORDS, EMBEDDING_DIM, weights=[embedding_matrix], trainable=False, name='glove_text_embedding')(input_text)
m1_layers = tf.keras.layers.Dropout(dropout_rate, name='dropout_multi_text_3')(m1_layers)
m1_layers = tf.keras.layers.Flatten(name='flatten_text')(m1_layers)
m1_layers = tf.keras.layers.Dropout(dropout_rate, name='dropout_multi_text_4')(m1_layers)
m1_layers = tf.keras.layers.Dense(512, activation='relu', name='dropout_multi_text_5')(m1_layers)
m1_layers = tf.keras.layers.Dropout(dropout_rate)(m1_layers)
m1_layers = tf.keras.layers.Dense(256,activation='relu', name='dense_1_text')(m1_layers)
m1_layers = tf.keras.layers.Dropout(dropout_rate, name='dropout_multi_text_1')(m1_layers)
m1_layers = tf.keras.layers.Dense(128,activation='relu', name='dense_2_text')(m1_layers)
m1_layers = tf.keras.layers.Dropout(dropout_rate, name='dropout_multi_text_2')(m1_layers)
m1_layers = tf.keras.layers.Dense(64,activation='relu', name='dense_3_text')(m1_layers)
if(count==1):
m1_layers = tf.keras.layers.Dense(2, activation='softmax', name='dense_output')(m1_layers)
model_1 = tf.keras.models.Model(inputs=input_text, outputs=m1_layers, name='texts_model')
mod_out.append(model_1.output)
mod_in.append(input_text)
if(sents==True):
input_sents = tf.keras.layers.Input(shape=(4096,),name="input_sents")
m1_layers = tf.keras.layers.Dense(1024, activation='relu')(input_sents)
m1_layers = tf.keras.layers.Dropout(0.2)(m1_layers)
m1_layers = tf.keras.layers.Dense(512, activation='relu')(m1_layers)
m1_layers = tf.keras.layers.Dropout(0.2)(m1_layers)
m1_layers = tf.keras.layers.Dense(128, activation='relu')(m1_layers)
m1_layers = tf.keras.layers.Dropout(0.4)(m1_layers)
m1_layers = tf.keras.layers.Dense(64, activation="relu")(m1_layers)
#m1_layers = tf.keras.layers.Dropout(0.2)(m1_layers)
#m1_layers = tf.keras.layers.Dense(32, activation="relu")(m1_layers)
#m1_layers = tf.keras.layers.Dropout(0.2)(m1_layers)
if(count==1):
m1_layers = tf.keras.layers.Dense(2, activation='softmax', name='dense_output')(m1_layers)
model_1 = tf.keras.models.Model(inputs=input_sents, outputs=m1_layers, name='sents_model')
mod_out.append(model_1.output)
mod_in.append(input_sents)
if topics==True:
input_topics = tf.keras.layers.Input(shape=(TOPICS_LEN,), name='input_topics')
#m2_layers = tf.keras.layers.Embedding(MAX_NB_WORDS, EMBEDDING_DIM, input_length=10, weights=[embedding_matrix], trainable=True, name='glove_topic_embedding')(input_topics)
#m2_layers = tf.keras.layers.Flatten(name='flatten_topics')(m2_layers)
m2_layers = tf.keras.layers.Dense(512,activation='relu', name='dense_1_topics')(input_topics)
m2_layers = tf.keras.layers.Dropout(0.2, name='dropout_multi_topic_1')(m2_layers)
m2_layers = tf.keras.layers.Dense(128,activation='relu', name='dense_2_topics')(m2_layers)
m2_layers = tf.keras.layers.Dropout(0.2, name='dropout_multi_topic_2')(m2_layers)
m2_layers = tf.keras.layers.Dense(64,activation='relu', name='dense_3_topics')(m2_layers)
if(count==1):
m2_layers = tf.keras.layers.Dense(2, activation='softmax', name='dense_output')(m2_layers)
model_2 = tf.keras.models.Model(inputs=input_topics, outputs=m2_layers, name='topics_model')
mod_out.append(model_2.output)
mod_in.append(input_topics)
if entities==True:
input_entities = tf.keras.layers.Input(shape=(ENTITIES_LEN,), name='input_entities')
m3_layers = tf.keras.layers.Embedding(MAX_NB_WORDS, EMBEDDING_DIM, input_length=ENTITIES_LEN, weights=[embedding_matrix], trainable=False, name='glove_entity_embedding')(input_entities)
#m3_layers= tf.keras.layers.Conv1D(32, 9, activation='relu', name='conv1d_1_ent')(m3_layers)
#m3_layers = tf.keras.layers.MaxPooling1D(4, name='maxpool1d_1_ent')(m3_layers)
#m3_layers = tf.keras.layers.Dropout(0.5, name='dropout_1_ent')(m3_layers)
#m3_layers = tf.keras.layers.Dropout(0.2)(m3_layers)
m3_layers = tf.keras.layers.Flatten(name='flatten_entities')(m3_layers)
#m3_layers = tf.keras.layers.Dense(1024,activation='relu', name='dense_3_entities_1')(m3_layers)
#m3_layers = tf.keras.layers.Dropout(0.4)(m3_layers)
m3_layers = tf.keras.layers.Dense(512,activation='relu', name='dense_3_entities_2')(m3_layers)
m3_layers = tf.keras.layers.Dropout(0.4)(m3_layers)
m3_layers = tf.keras.layers.Dense(128,activation='relu', kernel_regularizer="l1", name='dense_3_entities_5')(m3_layers)
m3_layers = tf.keras.layers.Dropout(0.3)(m3_layers)
m3_layers = tf.keras.layers.Dense(64,activation='relu', name='dense_3_entities_3')(m3_layers)
#kernel_regularizer=regularizers.l1_l2(l1=1e-5, l2=1e-4), bias_regularizer=regularizers.l2(1e-4), activity_regularizer=regularizers.l2(1e-5)
#m3_layers = tf.keras.layers.Dropout(0.2)(m3_layers)
if(count==1):
m3_layers = tf.keras.layers.Dense(2, activation='softmax', name='dense_output')(m3_layers)
model_3 = tf.keras.models.Model(inputs=input_entities, outputs=m3_layers)
mod_out.append(model_3.output)
mod_in.append(input_entities)
if triples==True:
input_triples = tf.keras.layers.Input(shape=(TRIPLES_LEN,), name='input_triples')
m4_layers = tf.keras.layers.Embedding(MAX_NB_WORDS, EMBEDDING_DIM, weights=[embedding_matrix], trainable=False, name='glove_triple_embedding')(input_triples)
#m4_layers= tf.keras.layers.Conv1D(32, 9, activation='relu', name='conv1d_1_triples')(m4_layers)
#m4_layers = tf.keras.layers.MaxPooling1D(4, name='maxpool1d_1_triples')(m4_layers)
m4_layers = tf.keras.layers.Dropout(0.2, name='dropout_1_triples')(m4_layers)
#m4_layers = tf.keras.layers.Dropout(0.2)(m4_layers)
m4_layers = tf.keras.layers.Flatten(name='flatten_triples')(m4_layers)
#m4_layers = tf.keras.layers.Dropout(0.4)(m4_layers)
#m4_layers = tf.keras.layers.Dense(1024,activation='relu', name='dense_4_triples_4')(m4_layers)
#m4_layers = tf.keras.layers.Dropout(0.5)(m4_layers)
m4_layers = tf.keras.layers.Dense(512,activation='relu', name='dense_4_triples_1')(m4_layers)
m4_layers = tf.keras.layers.Dropout(dropout_rate)(m4_layers)
m4_layers = tf.keras.layers.Dense(128,activation='relu',kernel_regularizer="l1" , name='dense_4_triples_2')(m4_layers)
m4_layers = tf.keras.layers.Dropout(dropout_rate)(m4_layers)
m4_layers = tf.keras.layers.Dense(64,activation='relu', name='dense_4_triples_3')(m4_layers)
#m4_layers = tf.keras.layers.Dropout(0.2)(m4_layers)
'''input_triples = tf.keras.layers.Input(shape=(None,), name='input_triples')
m4_layers = tf.keras.layers.Embedding(MAX_NB_WORDS, EMBEDDING_DIM, weights=[embedding_matrix], trainable=True, name='glove_triple_embedding')(input_triples)
m4_layers= tf.keras.layers.LSTM(256, kernel_regularizer="l1" ,name='lstm_1_triples')(m4_layers)
m4_layers= tf.keras.layers.Dropout(0.5)(m4_layers)'''
if(count==1):
m4_layers = tf.keras.layers.Dense(2, activation='softmax', name='dense_output')(m4_layers)
model_4 = tf.keras.models.Model(inputs=input_triples, outputs=m4_layers)
mod_out.append(model_4.output)
mod_in.append(input_triples)
if (count>1):
model_cat = tf.keras.layers.concatenate(mod_out)
model_cat = tf.keras.layers.Dense(32, activation="relu", name='dense_out')(model_cat)
model_cat = tf.keras.layers.Dropout(0.2)(model_cat)
model_cat = tf.keras.layers.Dense(2, activation='softmax', name='predictions')(model_cat)
model = tf.keras.models.Model(mod_in, model_cat, name='Model_Multi')
else:
if sents==True:
model=model_1
if text==True:
model=model_1
if topics==True:
model=model_2
if entities==True:
model=model_3
if triples==True:
model=model_4
optimiser = tf.keras.optimizers.Adam(learning_rate=learning_rate, epsilon=1e-08)
#ce = tf.keras.losses.BinaryCrossentropy()
ce = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
model.compile(loss=ce, optimizer=optimiser, metrics=['accuracy'])
#model.summary()
return model
def compute_metrics(model, X_test, Y_test):
yhat_probs = model.predict(X_test, verbose=0)
#yhat_classes = model.predict_classes(X_test, verbose=0)
Y_test_bin=[]
for y_bin in Y_test:
if (y_bin[0] == 1):
Y_test_bin.append(0)
else:
Y_test_bin.append(1)
#Y_test_bin = Y_test
yhat_classes = np.argmax(yhat_probs,axis=1)
print ("Accuracy|Precision|Recall|F1 score|Kappa|ROC AUC|")
# accuracy: (tp + tn) / (p + n)
accuracy = accuracy_score(Y_test_bin, yhat_classes)
#print('|%f|' % accuracy)
# precision tp / (tp + fp)
precision = precision_score(Y_test_bin, yhat_classes)
#print('%f|' % precision)
# recall: tp / (tp + fn)
recall = recall_score(Y_test_bin, yhat_classes)
#print('%f|' % recall)
# f1: 2 tp / (2 tp + fp + fn)
f1 = f1_score(Y_test_bin, yhat_classes)
#print('%f|' % f1)
# kappa
kappa = cohen_kappa_score(Y_test_bin, yhat_classes)
#print('%f|' % kappa)
# ROC AUC
auc = roc_auc_score(Y_test, yhat_probs)
print('%f\t%f\t%f\t%f\t%f\t%f' %(accuracy,precision,recall,f1,kappa,auc))
# confusion matrix
matrix = confusion_matrix(Y_test_bin, yhat_classes)
print(matrix)
print (Y_test_bin)
print (yhat_classes)
return ([accuracy, precision, recall, f1, kappa, auc], matrix)