-
Notifications
You must be signed in to change notification settings - Fork 19
/
Copy pathCNN_elmo.py
executable file
·160 lines (139 loc) · 6.27 KB
/
CNN_elmo.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
import ast
import numpy as np
from keras.preprocessing import sequence
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import StratifiedKFold
from keras.layers import Input, Flatten, Dense, Activation,Average
from keras.layers import Concatenate,Dropout,Conv1D,MaxPooling1D,BatchNormalization
from keras.models import Model
from keras import backend as K
from keras.callbacks import ModelCheckpoint
from argparse import ArgumentParser
def load_elmo(path, max_len=200):
'''
load ELMo embedding from tsv file.
:param path: tsv file path.
:param to_pickle: Convert elmo embeddings to .npy file, avoid read and pad every time.
:return: elmo embedding and its label.
'''
X = []
label = []
ids = []
i = 0
l_encoder = LabelEncoder()
with open(path, 'rb') as inf:
for line in inf:
gzip_fields = line.decode('utf-8').split('\t')
gzip_id = gzip_fields[0]
gzip_label = gzip_fields[1]
elmo_embd_str = gzip_fields[4].strip()
elmo_embd_list = ast.literal_eval(elmo_embd_str)
elmo_embd_array = np.array(elmo_embd_list)
padded_seq = sequence.pad_sequences([elmo_embd_array], maxlen=max_len, dtype='float32')[0]
X.append(padded_seq)
label.append(gzip_label)
ids.append(gzip_id)
i += 1
print(i)
Y = l_encoder.fit_transform(label)
return np.array(X), np.array(Y), np.array(ids)
def conv1d(max_len, embed_size):
'''
CNN without Batch Normalisation.
:param max_len: maximum sentence numbers, default=200
:param embed_size: ELMo embeddings dimension, default=1024
:return: CNN without BN model
'''
filter_sizes = [2, 3, 4, 5, 6]
num_filters = 128
drop = 0.5
inputs = Input(shape=(max_len,embed_size), dtype='float32')
conv_0 = Conv1D(num_filters, kernel_size=(filter_sizes[0]))(inputs)
act_0 = Activation('relu')(conv_0)
conv_1 = Conv1D(num_filters, kernel_size=(filter_sizes[1]))(inputs)
act_1 = Activation('relu')(conv_1)
conv_2 = Conv1D(num_filters, kernel_size=(filter_sizes[2]))(inputs)
act_2 = Activation('relu')(conv_2)
conv_3 = Conv1D(num_filters, kernel_size=(filter_sizes[3]))(inputs)
act_3 = Activation('relu')(conv_3)
conv_4 = Conv1D(num_filters, kernel_size=(filter_sizes[4]))(inputs)
act_4 = Activation('relu')(conv_4)
maxpool_0 = MaxPooling1D(pool_size=(max_len - filter_sizes[0]))(act_0)
maxpool_1 = MaxPooling1D(pool_size=(max_len - filter_sizes[1]))(act_1)
maxpool_2 = MaxPooling1D(pool_size=(max_len - filter_sizes[2]))(act_2)
maxpool_3 = MaxPooling1D(pool_size=(max_len - filter_sizes[3]))(act_3)
maxpool_4 = MaxPooling1D(pool_size=(max_len - filter_sizes[4]))(act_4)
concatenated_tensor = Concatenate()([maxpool_0, maxpool_1, maxpool_2, maxpool_3, maxpool_4])
flatten = Flatten()(concatenated_tensor)
dropout = Dropout(drop)(flatten)
output = Dense(units=1, activation='sigmoid')(dropout)
model = Model(inputs=inputs, outputs=output)
#model = multi_gpu_model(model, gpus=gpus)
model.summary()
model.compile(loss='binary_crossentropy', metrics=['acc'], optimizer='adam')
return model
def conv1d_BN(max_len, embed_size):
'''
CNN with Batch Normalisation.
:param max_len: maximum sentence numbers, default=200
:param embed_size: ELMo embeddings dimension, default=1024
:return: CNN with BN model
'''
filter_sizes = [2, 3, 4, 5, 6]
num_filters = 128
inputs = Input(shape=(max_len,embed_size), dtype='float32')
conv_0 = Conv1D(num_filters, kernel_size=(filter_sizes[0]))(inputs)
act_0 = Activation('relu')(conv_0)
bn_0 = BatchNormalization(momentum=0.7)(act_0)
conv_1 = Conv1D(num_filters, kernel_size=(filter_sizes[1]))(inputs)
act_1 = Activation('relu')(conv_1)
bn_1 = BatchNormalization(momentum=0.7)(act_1)
conv_2 = Conv1D(num_filters, kernel_size=(filter_sizes[2]))(inputs)
act_2 = Activation('relu')(conv_2)
bn_2 = BatchNormalization(momentum=0.7)(act_2)
conv_3 = Conv1D(num_filters, kernel_size=(filter_sizes[3]))(inputs)
act_3 = Activation('relu')(conv_3)
bn_3 = BatchNormalization(momentum=0.7)(act_3)
conv_4 = Conv1D(num_filters, kernel_size=(filter_sizes[4]))(inputs)
act_4 = Activation('relu')(conv_4)
bn_4 = BatchNormalization(momentum=0.7)(act_4)
maxpool_0 = MaxPooling1D(pool_size=(max_len - filter_sizes[0]))(bn_0)
maxpool_1 = MaxPooling1D(pool_size=(max_len - filter_sizes[1]))(bn_1)
maxpool_2 = MaxPooling1D(pool_size=(max_len - filter_sizes[2]))(bn_2)
maxpool_3 = MaxPooling1D(pool_size=(max_len - filter_sizes[3]))(bn_3)
maxpool_4 = MaxPooling1D(pool_size=(max_len - filter_sizes[4]))(bn_4)
concatenated_tensor = Concatenate()([maxpool_0, maxpool_1, maxpool_2, maxpool_3, maxpool_4])
flatten = Flatten()(concatenated_tensor)
output = Dense(units=1, activation='sigmoid')(flatten)
model = Model(inputs=inputs, outputs=output)
#model = multi_gpu_model(model, gpus=gpus)
model.summary()
model.compile(loss='binary_crossentropy', metrics=['acc'], optimizer='adam')
return model
parser = ArgumentParser()
parser.add_argument("inputTSV", help="Elmo format input file")
args = parser.parse_args()
seed = 7
max_len = 200
embed_size = 1024
x_data, y_data, ids = load_elmo(args.inputTSV, max_len=max_len)
# sk-learn provides 10-fold CV wrapper.
kfold = StratifiedKFold(n_splits=10, shuffle=True, random_state=seed)
# list of validation accuracy from each fold.
cvscores = []
# counter to tell which fold.
i = 0
for train, test in kfold.split(x_data, y_data):
i += 1
print("current fold is : %s " % i)
model = conv1d_BN(max_len, embed_size)
checkpoints = ModelCheckpoint(filepath='./saved_models/BNCNN_vacc{val_acc:.4f}_f%s_e{epoch:02d}.hdf5' % str(i),
verbose=1,monitor='val_acc', save_best_only=True)
history = model.fit(x_data[train],y_data[train],batch_size=32,verbose=1, epochs=30,
validation_data=[x_data[test],y_data[test]],callbacks=[checkpoints])
# use the last validation accuracy from the 30 epochs
his_val = history.history['val_acc'][-1]
cvscores.append(his_val)
# clear memory
K.clear_session()
print("Final score: %.4f%% (+/- %.4f%%)" % (np.mean(cvscores), np.std(cvscores)))