-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathbuild_model_for_hyperparameter_search.py
262 lines (201 loc) · 9.28 KB
/
build_model_for_hyperparameter_search.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
258
259
260
261
262
# -*- coding: utf-8 -*-
"""
Created on Sat Dec 7 21:24:08 2019
@author: zju
"""
import os
from math import log
import numpy as np
import tensorflow as tf
from tensorflow.keras.layers import *
from tensorflow.keras import Model
from tensorflow.keras import backend as K
from tensorflow.keras.optimizers import Adam
from sklearn.model_selection import KFold, ShuffleSplit
from build_my_layer import MyMaskCompute, MySpatialDropout1D
from utility import random_arr, array_split
from input_preprocess import preprocess
import warnings
import argparse
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
warnings.filterwarnings("ignore")
parser = argparse.ArgumentParser(description='train DeepTrio')
parser.add_argument('--interaction_data', required=True, type=str, help='configuration of the PPI file, which contains the protein 1 id, protein 2 id and the class they belong to, and they are splited by table key')
parser.add_argument('--sequence_data', required=True, type=str, help='configuration of the protein sequence database, which contains the protein id and its sequence, and they are splited by table key')
parser.add_argument('-e', '--epoch', default='100', type=str, help='the maximum number of epochs')
parser.add_argument('--cuda', default=False, action='store_true', help='Whether apply GPU to train the model')
static_args = parser.parse_args()
file_1_path = static_args.interaction_data
file_2_path = static_args.sequence_data
epoch_number = int(static_args.epoch)
if static_args.cuda:
print('using GPU\n')
else:
os.environ["CUDA_VISIBLE_DEVICES"] = ""
print('using CPU\n')
print('\nWelcome to use our tool')
print('\nVersion: 1.0.0')
print('\nAny problem, please contact [email protected]')
print('\nStart to process the raw data')
x_train_1, x_train_2, y_train, single_1, single_2, single_y, fix_text = preprocess(file_1_path, file_2_path)
random_arr(x_train_1)
random_arr(x_train_2)
random_arr(y_train)
random_arr(fix_text)
x_1, x_2, y_train, v_1, v_2, v_y, h_train, h_validation = array_split(10, x_train_1, x_train_2, y_train, single_1, single_2, single_y, fix_text)
print('\nStart to train DeepTrio model')
print('\nAfter training, you may select the best model manually according to the recording file')
def main(em_dim=15, sp_drop=0.005, kernel_rate_1=0.16, strides_rate_1=0.15, kernel_rate_2=0.14, strides_rate_2=0.25, filter_num_1=150, filter_num_2=175, con_drop=0.05, fn_drop_1=0.2, fn_drop_2=0.1, node_num=256, opti_switch=1):
if opti_switch == 0:
adam = Adam(amsgrad = False)
# print('^^^^^ False ^^^^^')
elif opti_switch == 1:
adam = Adam(amsgrad = True)
# print('^^^^^ True ^^^^^')
else:
raise Exception('The format is not in a right way')
main_input_a = Input(shape = (1500,), name = 'input_a')
main_input_b = Input(shape = (1500,), name = 'input_b')
embedding_layer = Embedding(25,int(em_dim),mask_zero=True)
embedded_a = embedding_layer(main_input_a)
embedded_b = embedding_layer(main_input_b)
masked_a = MyMaskCompute()(embedded_a)
masked_b = MyMaskCompute()(embedded_b)
drop_layer = MySpatialDropout1D(sp_drop)
dropped_1 = drop_layer(masked_a)
dropped_2 = drop_layer(masked_b)
tensor = []
for n in range(2,35):
if n <= 15:
conv_layer = Conv1D(filters= int(filter_num_1),
kernel_size = int(np.ceil(kernel_rate_1 * n**2)),
padding = 'valid',
activation = 'relu',
use_bias= False,
strides = int(np.ceil(strides_rate_1*(n-1))))
else:
conv_layer = Conv1D(filters= int(filter_num_2),
kernel_size = int(np.ceil(kernel_rate_2 * n**2)),
padding = 'valid',
activation = 'relu',
use_bias= False,
strides = int(np.ceil(strides_rate_2*(n-1))))
conv_out_1 = conv_layer(dropped_1)
conv_out_2 = conv_layer(dropped_2)
conv_out_1 = SpatialDropout1D(con_drop)(conv_out_1)
conv_out_2 = SpatialDropout1D(con_drop)(conv_out_2)
max_layer = MaxPooling1D(pool_size=int(conv_layer.output_shape[1]))
pool_out_1 = max_layer(conv_out_1)
pool_out_2 = max_layer(conv_out_2)
pool_out = pool_out_1 + pool_out_2
flat_out = Flatten()(pool_out)
tensor.append(flat_out)
concatenated = Concatenate()(tensor)
x = Dropout(fn_drop_1)(concatenated)
x = Dense(int(node_num))(x)
x = Dropout(fn_drop_2)(x)
x = Activation('relu')(x)
x = Dense(3)(x)
main_output = Activation('softmax', name = 'out')(x)
model = Model(inputs = [main_input_a,main_input_b], outputs = main_output)
model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy'])
record_min = 0
for n in range(epoch_number):
history_model = model.fit([x_1[0],x_2[0]], y_train[0], batch_size=256, epochs=1, shuffle=True, validation_data=([v_1[0],v_2[0]], v_y[0]))
if history_model.history['val_accuracy'][0] > record_min:
record_min = history_model.history['val_accuracy'][0]
model.save('DeepTrio_search_' + str(yy) + '.h5') ##########
print(str(record_min)+'\taccuracy_model_saved')
with open('search_log.txt', 'a') as log_text:
log_text.write('cycle: ' + str(yy) + '\n')
log_text.write('em_dim: ' + str(em_dim) + '\n')
log_text.write('sp_drop: ' + str(sp_drop) + '\n')
log_text.write('kernel_rate_1: ' + str(kernel_rate_1) + '\n')
log_text.write('strides_rate_1: ' + str(strides_rate_1) + '\n')
log_text.write('kernel_rate_2: ' + str(kernel_rate_2) + '\n')
log_text.write('strides_rate_2: ' + str(strides_rate_2) + '\n')
log_text.write('filter_num_1: ' + str(filter_num_1) + '\n')
log_text.write('filter_num_2: ' + str(filter_num_2) + '\n')
log_text.write('con_drop: ' + str(con_drop) + '\n')
log_text.write('fn_drop_1: ' + str(fn_drop_1) + '\n')
log_text.write('fn_drop_2: ' + str(fn_drop_2) + '\n')
log_text.write('node_num: ' + str(node_num) + '\n')
log_text.write('opti_switch: ' + str(opti_switch) + '\n')
log_text.write('accuracy: ' + str(record_min) + '\n')
log_text.write('-----\n')
return record_min
bounds = [
#Discrete
{'name':'em_dim','type':'discrete','domain':(10, 15, 20)},
{'name':'sp_drop','type':'discrete','domain':(0.005, 0.01)},
{'name':'kernel_rate_1','type':'discrete','domain':(0.12, 0.14, 0.16)},
{'name':'strides_rate_1','type':'discrete','domain':(0.15, 0.2, 0.25)},
{'name':'kernel_rate_2','type':'discrete','domain':(0.1, 0.12, 0.14)},
{'name':'strides_rate_2','type':'discrete','domain':(0.25, 0.3, 0.35)},
{'name':'filter_num_1','type':'discrete','domain':(100, 125, 150)},
{'name':'filter_num_2','type':'discrete','domain':(150, 175)},
{'name':'con_drop', 'type': 'discrete','domain': (0.05, 0.1, 0.15)},
{'name':'fn_drop_1', 'type': 'discrete','domain': (0.1, 0.2)},
{'name':'fn_drop_2', 'type': 'discrete','domain': (0.1, 0.2)},
{'name':'node_num', 'type': 'discrete','domain': (128, 256)},
#Categorical
{'name':'opti_switch', 'type': 'categorical','domain': (0, 1)}
]
from dotmap import DotMap
def search_param(x):
opt= DotMap()
opt.em_dim = float(x[:, 0])
opt.sp_drop = float(x[:, 1])
opt.kernel_rate_1 = float(x[:, 2])
opt.strides_rate_1 = float(x[:, 3])
opt.kernel_rate_2 = float(x[:, 4])
opt.strides_rate_2 = float(x[:, 5])
opt.filter_num_1 = float(x[:, 6])
opt.filter_num_2 = float(x[:, 7])
opt.con_drop = float(x[:, 8])
opt.fn_drop_1 = float(x[:, 9])
opt.fn_drop_2 = float(x[:, 10])
opt.node_num = float(x[:, 11])
opt.opti_switch = int(x[:,12])
return opt
yy = 0
def f(x):
global yy
local_yy = yy
local_yy += 1
yy = local_yy
opt = search_param(x)
param = {
'em_dim':opt.em_dim,
'sp_drop':opt.sp_drop,
'kernel_rate_1':opt.kernel_rate_1,
'strides_rate_1':opt.strides_rate_1,
'kernel_rate_2':opt.kernel_rate_2,
'strides_rate_2':opt.strides_rate_2,
'filter_num_1':opt.filter_num_1,
'filter_num_2':opt.filter_num_2,
'con_drop':opt.con_drop,
'fn_drop_1':opt.fn_drop_1,
'fn_drop_2':opt.fn_drop_2,
'node_num':opt.node_num,
'opti_switch':opt.opti_switch
}
result = main(**param)
evaluation = 1 - result
# with open('search_log.txt', 'a') as log_text:
# log_text.write('evaluation: ' + str(evaluation) + '\n')
# log_text.write('---------\n')
print('cycle: ' + str(yy))
print('evaluation: ' + str(evaluation))
return evaluation
import GPy
import GPyOpt
opt_model = GPyOpt.methods.BayesianOptimization(f=f, domain=bounds, initial_design_numdata=10)
opt_model.run_optimization(max_iter=50)
with open('search_log.txt', 'a') as log_text:
log_text.write('result: \n')
for i,v in enumerate(bounds):
name = v['name']
log_text.write('parameter {}: {}\n'.format(name,opt_model.x_opt[i]))
log_text.write('evaluation: ' + str(1 - opt_model.fx_opt) + '\n')
print('Congratulations, the training is complete')