/
share.py
207 lines (199 loc) · 9.89 KB
/
share.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
import numpy as np
from keras.preprocessing import sequence
from keras.utils import np_utils
from keras.models import Sequential, Graph
from keras.layers.core import Dense, Dropout, Activation
from keras.layers.embeddings import Embedding
from keras.layers.recurrent import LSTM
from keras.layers.normalization import BatchNormalization
from keras.optimizers import Adam
from keras.constraints import maxnorm
from keras import backend as K
from keras.callbacks import ModelCheckpoint
import theano
import theano.tensor as T
import ast
"""
def loss(y_true, y_pred):
global output_feature;
y_true_positive=y_true[:,:output_feature]
y_true_negative=y_true[:,output_feature:]
y_pred = K.l2_normalize(y_pred, axis=1)
y_true_positive = K.l2_normalize(y_true_positive, axis=1)
y_true_negative = K.l2_normalize(y_true_negative, axis=1)
return K.maximum(K.sum(y_pred*y_true_negative, axis=1) - K.sum(y_pred*y_true_positive, axis=1)+0.6, 0.)
"""
"""
def hinge(y_true, y_pred):
return K.mean(K.maximum(1. - y_true * y_pred, 0.), axis=-1)
def loss(y_true, y_pred):
global output_feature;
y_true_positive=y_true[:,:output_feature]
y_true_negative=y_true[:,output_feature:]
return 0.1*cosine_proximity(y_true_negative, y_pred)-cosine_proximity(y_true_positive, y_pred)
"""
"""
def cosine_proximity(y_true, y_pred):
assert K.ndim(y_true) == 2
assert K.ndim(y_pred) == 2
y_true = K.l2_normalize(y_true, axis=1)
y_pred = K.l2_normalize(y_pred, axis=1)
return -K.mean(y_true * y_pred, axis=1)
"""
"""
def loss(y_true, y_pred):
global output_feature;
y_pred_positive=y_pred[:,:output_feature]
y_pred_label=y_pred[:,output_feature:2*output_feature]
y_pred_negative=y_pred[:,2*output_feature:]
l=K.mean(y_true-y_true, axis=1)
return -cosine_proximity(y_pred_positive, y_pred_negative)+cosine_proximity(y_pred_positive, y_pred_label)+l
#return K.sum(y_pred_positive*y_true, axis=1)
"""
def euclidean(v1, v2):
return np.sqrt(np.sum((v1-v2)*(v1-v2)))
def cosine_proximity(y_true, y_pred):
assert K.ndim(y_true) == 2
assert K.ndim(y_pred) == 2
y_true = K.l2_normalize(y_true, axis=1)
y_pred = K.l2_normalize(y_pred, axis=1)
return K.sum(y_true * y_pred, axis=1)
def mean_squared_error(y_true, y_pred):
return K.square(K.sum((y_pred - y_true)*(y_pred - y_true), axis=-1))
def loss(y_true, y_pred):
global output_feature;
y_pred_positive=y_pred[:,:output_feature]
y_pred_label=y_pred[:,output_feature:2*output_feature]
y_pred_negative1=y_pred[:,2*output_feature:3*output_feature]
y_pred_negative2=y_pred[:,3*output_feature:4*output_feature]
y_pred_negative3=y_pred[:,4*output_feature:5*output_feature]
y_pred_negative4=y_pred[:,5*output_feature:6*output_feature]
y_pred_negative5=y_pred[:,6*output_feature:7*output_feature]
y_pred_negative6=y_pred[:,7*output_feature:8*output_feature]
l=K.min(y_true-y_true, axis=1)
return K.maximum(cosine_proximity(y_pred_positive, y_pred_negative1)-cosine_proximity(y_pred_positive, y_pred_label)+0.3, 0.)\
+K.maximum(cosine_proximity(y_pred_positive, y_pred_negative2)-cosine_proximity(y_pred_positive, y_pred_label)+0.3, 0.)\
+K.maximum(cosine_proximity(y_pred_positive, y_pred_negative3)-cosine_proximity(y_pred_positive, y_pred_label)+0.3, 0.)\
+K.maximum(cosine_proximity(y_pred_positive, y_pred_negative4)-cosine_proximity(y_pred_positive, y_pred_label)+0.3, 0.)\
+K.maximum(cosine_proximity(y_pred_positive, y_pred_negative5)-cosine_proximity(y_pred_positive, y_pred_label)+0.3, 0.)\
+K.maximum(cosine_proximity(y_pred_positive, y_pred_negative6)-cosine_proximity(y_pred_positive, y_pred_label)+0.3, 0.)\
+K.maximum(0.98-cosine_proximity(y_pred_positive, y_pred_label), 0.)+l
"""
def loss(y_true, y_pred):
global output_feature;
y_pred_positive=y_pred[:,:output_feature]
y_pred_label=y_pred[:,output_feature:2*output_feature]
y_pred_negative=y_pred[:,2*output_feature:]
l=K.min(y_true-y_true, axis=1)
return cosine_proximity(y_pred_positive, y_pred_negative)-cosine_proximity(y_pred_positive, y_pred_label)+l
#return K.maximum(cosine_proximity(y_pred_positive, y_pred_negative)-cosine_proximity(y_pred_positive, y_pred_label)+0.8, 0.)+l
#return -cosine_proximity(y_pred_positive, y_pred_label)+l
"""
output_feature=300
input_feature=300
length=8
batch_size = 128
dictionary_size=6001
features=[]
labels=[]
labels_negative1=[]
labels_negative2=[]
labels_negative3=[]
labels_negative4=[]
labels_negative5=[]
labels_negative6=[]
answers=[]
embedding_size=300
f=open('vectorpair.txt')
#f=open('vectorpair2.txt')
for line in f:
record=line.split('|||')
if len(record) != length:
print((record))
continue
features+=[ast.literal_eval(record[0])]
labels+=[ast.literal_eval(record[1])]
labels_negative1+=[ast.literal_eval(record[2])]
labels_negative2+=[ast.literal_eval(record[3])]
labels_negative3+=[ast.literal_eval(record[4])]
labels_negative4+=[ast.literal_eval(record[5])]
labels_negative5+=[ast.literal_eval(record[6])]
labels_negative6+=[ast.literal_eval(record[7])]
f=open('vectorid.txt')
#f=open('vectorid2.txt')
for line in f:
answers+=[ast.literal_eval(line)]
features=np.array(features)
labels=np.array(labels)
labels_negative1=np.array(labels_negative1)
labels_negative2=np.array(labels_negative2)
labels_negative3=np.array(labels_negative3)
labels_negative4=np.array(labels_negative4)
labels_negative5=np.array(labels_negative5)
labels_negative6=np.array(labels_negative6)
answers=np.array(answers)
features = sequence.pad_sequences(features, maxlen=input_feature)
labels=sequence.pad_sequences(labels, maxlen=input_feature)
labels_negative1=sequence.pad_sequences(labels_negative1, maxlen=input_feature)
labels_negative2=sequence.pad_sequences(labels_negative2, maxlen=input_feature)
labels_negative3=sequence.pad_sequences(labels_negative3, maxlen=input_feature)
labels_negative4=sequence.pad_sequences(labels_negative4, maxlen=input_feature)
labels_negative5=sequence.pad_sequences(labels_negative5, maxlen=input_feature)
labels_negative6=sequence.pad_sequences(labels_negative6, maxlen=input_feature)
answers=sequence.pad_sequences(answers, maxlen=input_feature)
print('Build model...')
model = Graph()
model.add_input(name='input1', input_shape=(input_feature,), dtype='float' )
model.inputs["input1"].input = T.imatrix()
model.add_input(name='input2', input_shape=(input_feature,), dtype='float')
model.inputs["input2"].input = T.imatrix()
model.add_input(name='input3', input_shape=(input_feature,), dtype='float')
model.inputs["input3"].input = T.imatrix()
model.add_input(name='input4', input_shape=(input_feature,), dtype='float')
model.inputs["input4"].input = T.imatrix()
model.add_input(name='input5', input_shape=(input_feature,), dtype='float')
model.inputs["input5"].input = T.imatrix()
model.add_input(name='input6', input_shape=(input_feature,), dtype='float')
model.inputs["input6"].input = T.imatrix()
model.add_input(name='input7', input_shape=(input_feature,), dtype='float')
model.inputs["input7"].input = T.imatrix()
model.add_input(name='input8', input_shape=(input_feature,), dtype='float')
model.inputs["input8"].input = T.imatrix()
shared_dense1=Embedding(dictionary_size, embedding_size, input_length=input_feature,init='he_uniform')
shared_dense2=LSTM(embedding_size,init='he_uniform')
shared_dense3=BatchNormalization()
shared_dense4=Dense(50,W_constraint = maxnorm(1.2))
shared_dense5=Dense(50,W_constraint = maxnorm(1.2))
shared_dense6=Dense(50,W_constraint = maxnorm(1.2))
shared_dense7=Dense(output_feature,W_constraint = maxnorm(1.2))
model.add_shared_node(shared_dense1, name="shared_dense1", inputs=['input1','input2','input3','input4','input5','input6','input7','input8'], merge_mode=None)
model.add_shared_node(shared_dense2, name="shared_dense2", inputs=["shared_dense1"], merge_mode=None)
model.add_shared_node(shared_dense3, name="shared_dense3", inputs=["shared_dense2"], merge_mode=None)
model.add_shared_node(shared_dense4, name="shared_dense4", inputs=["shared_dense3"], merge_mode=None)
model.add_shared_node(shared_dense5, name="shared_dense5", inputs=["shared_dense4"], merge_mode=None)
model.add_shared_node(shared_dense6, name="shared_dense6", inputs=["shared_dense5"], merge_mode=None)
model.add_shared_node(shared_dense7, name="shared_dense7", inputs=["shared_dense6"], merge_mode='concat')
model.add_node(Activation('softplus'), name='out', input='shared_dense7')
model.add_output(name='output', input='out')
adam=Adam(clipnorm=0.09)
model.compile(adam, {'output': loss})
print("Train...")
checkpointer = ModelCheckpoint(filepath="./weight/weights.{epoch:02d}.hdf5", verbose=1, save_best_only=False)
model.fit({'input1': features, 'input2': labels, 'input3': labels_negative1,'input4': labels_negative2,'input5': labels_negative3,'input6': labels_negative4,'input7': labels_negative5,'input8': labels_negative6, 'output': features},
batch_size=batch_size, callbacks=[checkpointer],
nb_epoch=30)
result=np.array(model.predict({'input1': features, 'input2': features, 'input3': features, 'input4': features, 'input5': features, 'input6': features, 'input7': features, 'input8': features}, batch_size=batch_size)['output'])
a=result[:, :output_feature]
result=np.array(model.predict({'input1': answers, 'input2': answers, 'input3': answers, 'input4': answers, 'input5': answers, 'input6': answers, 'input7': answers, 'input8': answers}, batch_size=batch_size)['output'])
b=result[:, :output_feature]
with open('result1.txt', 'w') as ff:
for i in a:
ff.write(str(i.tolist()))
ff.write('\n')
with open('result2.txt', 'w') as ff:
for i in b:
ff.write(str(i.tolist()))
ff.write('\n')
json_string = model.to_json()
open('my_model_architecture.json', 'w').write(json_string)
model.save_weights('my_model_weights.h5', overwrite=True)