-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathclassifier_models.py
259 lines (229 loc) · 12.7 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
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
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
from tensorflow import keras
from bert_util import BERT_LAYER_LINK
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
from util import MAX_NB_WORDS, EMBEDDING_DIM, TEXT_LEN, TOPICS_LEN, ENTITIES_LEN, TRIPLES_LEN, BERT_SEQ_LEN
from util import SENTENCE_EMBEDDINGS, GLOVE_EMBEDDINGS, BERT
def model_making(count, embedding_matrix, topics=False, entities=False, triples=False, text=False, fine_tune=False, embedding='glove', num_labels=2):
"""
This function creates the models based on the input provided by the user. This function is usually called from the generate_model function from train.py
"""
learning_rate = 2e-4
if(embedding==BERT):
learning_rate = 2e-5
mod_out=[]
mod_in=[]
dropout_rate = 0.3
dropout_rate_2 = 0.2
if (text==True):
if(embedding==GLOVE_EMBEDDINGS):
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=fine_tune, 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.Dense(512, activation='relu', name='dropout_multi_text_5')(m1_layers)
m1_layers = tf.keras.layers.Dropout(dropout_rate, name='dropout_multi_text_2')(m1_layers)
m1_layers = tf.keras.layers.Dense(100,activation='relu', name='dense_3_text')(m1_layers)
if(count==1):
m1_layers = tf.keras.layers.Dense(num_labels, 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)
elif(embedding==SENTENCE_EMBEDDINGS):
input_sents = tf.keras.layers.Input(shape=(4096,),name="input_sents")
m1_layers = tf.keras.layers.Dense(1024, activation='relu', name='dense_1_sents')(input_sents)
m1_layers = tf.keras.layers.Dropout(dropout_rate)(m1_layers)
m1_layers = tf.keras.layers.Dense(512, activation="relu")(m1_layers)
if(count==1):
m1_layers = tf.keras.layers.Dense(num_labels, 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)
elif(embedding==BERT):
input_word_ids = tf.keras.layers.Input(shape=(BERT_SEQ_LEN,), dtype=tf.int32,name="input_word_ids")
input_mask = tf.keras.layers.Input(shape=(BERT_SEQ_LEN,), dtype=tf.int32,name="input_mask")
segment_ids = tf.keras.layers.Input(shape=(BERT_SEQ_LEN,), dtype=tf.int32,name="segment_ids")
bert_text_inputs = [input_word_ids, input_mask, segment_ids]
bert_layer=hub.KerasLayer(BERT_LAYER_LINK,trainable=fine_tune, name="bert_layer_text")
pooled_output, sequence_output = bert_layer(bert_text_inputs)
m1_layers = tf.keras.layers.GlobalAveragePooling1D()(sequence_output)
m1_layers = tf.keras.layers.Dense(100, activation="relu")(m1_layers)
if(count==1):
m1_layers = tf.keras.layers.Dense(num_labels, activation='softmax', name='dense_output')(m1_layers)
model_1 = tf.keras.models.Model(inputs=bert_text_inputs, outputs=m1_layers, name='bert_text_model')
mod_out.append(model_1.output)
mod_in.append(bert_text_inputs)
if topics==True:
input_topics = tf.keras.layers.Input(shape=(TOPICS_LEN,), name='input_topics')
if(embedding==BERT):
m2_layers = tf.keras.layers.Dense(100,activation='relu', name='dense_2_topics')(input_topics)
else:
m2_layers = tf.keras.layers.Dropout(dropout_rate, name='dropout_multi_topic_4')(input_topics)
m2_layers = tf.keras.layers.Dense(512,activation='relu', name='dense_1_topics')(m2_layers)
m2_layers = tf.keras.layers.Dense(100,activation='relu', name='dense_2_topics')(m2_layers)
if(count==1):
m2_layers = tf.keras.layers.Dense(num_labels, 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=fine_tune, name='glove_entity_embedding')(input_entities)
m3_layers = tf.keras.layers.Dropout(dropout_rate, name='dropout_1_entities')(m3_layers)
m3_layers = tf.keras.layers.Flatten()(m3_layers)
m3_layers = tf.keras.layers.Dropout(dropout_rate)(m3_layers)
m3_layers = tf.keras.layers.Dense(512,activation='relu', name='dense_3_entities_2')(m3_layers)
m3_layers = tf.keras.layers.Dropout(dropout_rate)(m3_layers)
m3_layers = tf.keras.layers.Dense(100,activation='relu', name='dense_3_entities_4')(m3_layers)
if(count==1):
m3_layers = tf.keras.layers.Dense(num_labels, 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:
if(embedding==SENTENCE_EMBEDDINGS):
input_triples = tf.keras.layers.Input(shape=(4096,),name="input_triples")
m4_layers = tf.keras.layers.Dense(100,activation='relu', name='dense_4_triples_1')(input_triples)
#m4_layers = tf.keras.layers.Dropout(dropout_rate)(m4_layers)
m4_layers = tf.keras.layers.Dense(10,activation='relu', name='dense_4_triples_3')(m4_layers)
if(count==1):
m4_layers = tf.keras.layers.Dense(num_labels, 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)
elif(embedding==GLOVE_EMBEDDINGS):
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=fine_tune, 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.GlobalAveragePooling1D()(m4_layers)
m4_layers = tf.keras.layers.Dropout(dropout_rate, name='dropout_1_triples')(m4_layers)
m4_layers = tf.keras.layers.Flatten()(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(100,activation='relu', name='dense_4_triples_3')(m4_layers)
if(count==1):
m4_layers = tf.keras.layers.Dense(num_labels, 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)
elif (embedding==BERT):
input_word_ids = tf.keras.layers.Input(shape=(BERT_SEQ_LEN,), dtype=tf.int32,name="triple_input_word_ids")
input_mask = tf.keras.layers.Input(shape=(BERT_SEQ_LEN,), dtype=tf.int32,name="triple_input_mask")
segment_ids = tf.keras.layers.Input(shape=(BERT_SEQ_LEN,), dtype=tf.int32,name="triple_segment_ids")
bert_triple_inputs = [input_word_ids, input_mask, segment_ids]
bert_layer=hub.KerasLayer(BERT_LAYER_LINK,trainable=fine_tune, name="bert_layer_triples")
pooled_output, sequence_output = bert_layer(bert_triple_inputs)
m4_layers = tf.keras.layers.GlobalAveragePooling1D()(sequence_output)
m4_layers = tf.keras.layers.Dense(100, activation="relu")(m4_layers)
if(count==1):
m4_layers = tf.keras.layers.Dense(num_labels, activation='softmax', name='dense_output')(m4_layers)
model_4 = tf.keras.models.Model(inputs=bert_triple_inputs, outputs=m4_layers, name='bert_triple_model')
mod_out.append(model_4.output)
mod_in.append(bert_triple_inputs)
if (count>1):
model_cat = tf.keras.layers.Concatenate()(mod_out)
#model_cat = tf.keras.layers.Dense(512,activation='relu', name='dense_1_cat')(model_cat)
#model_cat = tf.keras.layers.Dense(100,activation='relu', name='dense_2_cat')(model_cat)
#model_cat = tf.keras.layers.Dropout(dropout_rate)(model_cat)
#model_cat = tf.keras.layers.Dense(100, activation="relu", name='dense_out')(model_cat)
#if(embedding==GLOVE_EMBEDDINGS):
# model_cat = tf.keras.layers.Flatten(name='flatten_layers')(model_cat)
model_cat = tf.keras.layers.Dense(num_labels, activation='softmax', name='predictions')(model_cat)
model = tf.keras.models.Model(mod_in, model_cat, name='Model_Multi')
else:
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.SGD(learning_rate=learning_rate)
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, name='text'):
"""
This function measures the model performance by calculating model accuracy, F1 score, precision, recall, AUC.
In case of a multi class classification - we measure accuracy, F1-Macro, F1-Micro, Weighted F-score
"""
yhat_probs = model.predict(X_test, verbose=0)
#yhat_classes = model.predict_classes(X_test, verbose=0)
Y_test_bin=[]
text=""
for y_bin in Y_test:
index = np.where(y_bin==1)
Y_test_bin.append(index[0][0])
#Y_test_bin = Y_test
yhat_classes = np.argmax(yhat_probs,axis=1)
num_labels=len(yhat_probs[0])
#print (Y_test_bin)
#print (yhat_classes)
# accuracy: (tp + tn) / (p + n)
accuracy = accuracy_score(Y_test_bin, yhat_classes)
if(num_labels==2):
#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)
res_text = "Accuracy|Precision|Recall|F1 score|Kappa|ROC AUC|"
#print (res_text)
#print('%f\t%f\t%f\t%f\t%f\t%f' %(accuracy,precision,recall,f1,kappa,auc))
res_text_2=str(accuracy)+"|"+str(precision)+"|"+str(recall)+"|"+str(f1)+"|"+str(kappa)+"|"+str(auc)
return_metrics=[accuracy, precision, recall, f1, kappa, auc]
else:
f1_macro = f1_score(Y_test_bin, yhat_classes, average='macro')
f1_micro = f1_score(Y_test_bin, yhat_classes, average='micro')
f1_weighted = f1_score(Y_test_bin, yhat_classes, average='weighted')
res_text = "Accuracy|F1-Macro|F1-Micro|F1-Weighted|"
#print (res_text)
res_text_2=str(accuracy)+"|"+str(f1_macro)+"|"+str(f1_micro)+"|"+str(f1_weighted)
#print('%f\t%f\t%f\t%f' %(accuracy,f1_macro,f1_micro,f1_weighted))
return_metrics=[accuracy, f1_macro,f1_micro,f1_weighted]
# confusion matrix
matrix = confusion_matrix(Y_test_bin, yhat_classes)
print(matrix)
#print (Y_test_bin)
#print (yhat_classes)
text+=res_text+"\n"
text+=res_text_2+"\n\n"
for res in Y_test_bin:
text +=str(res)+","
text=text[:-1]
text=text+"\n"
for res in yhat_classes:
text +=str(res)+","
text=text[:-1]
text=text+"\n\n"
if(num_labels==2):
target_names=['Relevant', 'Irrelevant']
text+=classification_report(Y_test_bin, yhat_classes, target_names=target_names)
f = open("results_2/"+name+".txt", "w")
f.write(text)
f.close()
return (return_metrics, matrix)