-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathdeep_convolutional_representation.py
382 lines (240 loc) · 11 KB
/
deep_convolutional_representation.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
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
'''
Create CNN based on ResNet-50 to generate flatten representation of the 2D feature matrix
'''
import numpy as np
import pandas as pd
from pyparsing import And
import os
from tensorflow.keras import activations, Input
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.models import Model, load_model
from tensorflow.keras.layers import Conv2D, BatchNormalization, Activation, Add, MaxPooling2D, ZeroPadding2D, AveragePooling2D
from tensorflow.keras.layers import Dense, Dropout, Activation, Flatten
from tensorflow.keras.regularizers import l2
from tensorflow.keras.optimizers.schedules import ExponentialDecay
from tensorflow.keras.callbacks import ReduceLROnPlateau, EarlyStopping, ModelCheckpoint, Callback
class prepare():
'''
Class used to create CNN which take SMILES feature matrice as inputs
args:
(1) path_to_save (type:str) - location to save new data files
'''
def __init__(self, path_to_save: str, *args, **kwargs):
self.path_to_save = path_to_save
os.chdir(self.path_to_save)
def identity_block(self, x, filters):
'''
Identity block
'''
x_skip = x
f1, f2 = filters
# First block
x = Conv2D(f1, kernel_size=(1, 1), strides=(1, 1), padding='valid', kernel_regularizer=l2(0.001))(x)
x = BatchNormalization()(x)
x = Activation(activations.relu)(x)
# Second block
x = Conv2D(f1, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(0.001))(x)
x = BatchNormalization()(x)
x = Activation(activations.relu)(x)
# Third block
x = Conv2D(f2, kernel_size=(1, 1), strides=(1, 1), padding='valid', kernel_regularizer=l2(0.001))(x)
x = BatchNormalization()(x)
# Skip connection addition
x = Add()([x, x_skip])
x = Activation(activations.relu)(x)
return x
def convolutional_block(self, x, s, filters):
'''
Convolutional block
'''
x_skip = x
f1, f2 = filters
# First block
x = Conv2D(f1, kernel_size=(1, 1), strides=(s, s), padding='valid', kernel_regularizer=l2(0.001))(x)
x = BatchNormalization()(x)
x = Activation(activations.relu)(x)
# Second block
x = Conv2D(f1, kernel_size=(3, 3), strides=(1, 1), padding='same', kernel_regularizer=l2(0.001))(x)
x = BatchNormalization()(x)
x = Activation(activations.relu)(x)
# Third block
x = Conv2D(f2, kernel_size=(1, 1), strides=(1, 1), padding='valid', kernel_regularizer=l2(0.001))(x)
x = BatchNormalization()(x)
# Skip connection
x_skip = Conv2D(f2, kernel_size=(1, 1), strides=(s, s), padding='valid', kernel_regularizer=l2(0.001))(x_skip)
x_skip = BatchNormalization()(x_skip)
# Skip connection addition
x = Add()([x, x_skip])
x = Activation(activations.relu)(x)
return x
def create_model(self, x_train, y_train, *args, **kwargs):
'''
Create deep model
'''
init_kernel_size = kwargs.get('init_kernel_size')
input_tensor = Input(shape=(x_train.shape[1], x_train.shape[2], 1))
x = ZeroPadding2D(padding=(3, 3))(input_tensor)
# Input stage
if init_kernel_size is None:
x = Conv2D(64, kernel_size=(3, 3), strides=(2, 2))(x)
elif init_kernel_size is not None:
x = Conv2D(64, kernel_size=init_kernel_size, strides=(2, 2))(x)
x = BatchNormalization()(x)
x = Activation(activations.relu)(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
# First block
x = self.convolutional_block(x, s=1, filters=(64, 256))
# Second block
x = self.convolutional_block(x, s=2, filters=(128, 512))
x = self.identity_block(x, filters=(128, 512))
x = self.identity_block(x, filters=(128, 512))
# Third block
x = self.convolutional_block(x, s=2, filters=(256, 1024))
x = self.identity_block(x, filters=(256, 1024))
x = self.identity_block(x, filters=(256, 1024))
x = self.identity_block(x, filters=(256, 1024))
# Fourth block
x = self.convolutional_block(x, s=2, filters=(512, 2048))
# Average pooling & dense connection
x = AveragePooling2D((2, 2), padding='same')(x)
x = Flatten()(x)
x = Dense(512, activation='relu', name='deep_representation')(x)
output_tensor = Dense(1, activation='linear', name='output')(x)
# Define the model
model = Model(inputs=input_tensor, outputs=output_tensor, name='Deep_CNN')
model.summary()
lr_schedule = ExponentialDecay(
initial_learning_rate=1e-2,
decay_steps=10000,
decay_rate=0.9
)
model.compile(
optimizer=Adam(lr_schedule),
#optimizer=Adam(1e-2),
loss='mse',
metrics=['mse', 'mae']
)
return model
def create_simpler_model(self, x_train, y_train , *args, **kwargs):
'''
Create model
'''
init_kernel_size = kwargs.get('init_kernel_size') # saved weight
input_tensor = Input(shape=(x_train.shape[1], x_train.shape[2], 1))
x = ZeroPadding2D(padding=(3, 3))(input_tensor)
# 1st stage
# here we perform maxpooling
if init_kernel_size is None:
x = Conv2D(64, kernel_size=(3, 3), strides=(2, 2))(x)
elif init_kernel_size is not None:
x = Conv2D(64, kernel_size=init_kernel_size, strides=(2, 2))(x)
x = BatchNormalization()(x)
x = Activation(activations.relu)(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
#2nd stage
# frm here on only conv block and identity block, no pooling
x = self.convolutional_block(x, s=1, filters=(64, 256))
# 3rd stage
x = self.convolutional_block(x, s=2, filters=(128, 512))
# 4th stage
x = self.convolutional_block(x, s=2, filters=(256, 1024))
# 5th stage
x = self.convolutional_block(x, s=2, filters=(512, 2048))
# ends with average pooling and dense connection
x = AveragePooling2D((2, 2), padding='same')(x)
x = Flatten()(x)
x = Dense(512, activation='relu')(x)
output_tensor = Dense(1, activation='linear')(x)
# define the model
model = Model(inputs=input_tensor, outputs=output_tensor, name='CNN')
#model.summary()
lr_schedule = ExponentialDecay(
initial_learning_rate=1e-2,
decay_steps=10000,
decay_rate=0.9
)
model.compile(
optimizer=Adam(lr_schedule),
#optimizer=Adam(1e-2),
loss='mse',
metrics=['mse', 'mae']
)
return model
def resnet(
self,
x_train: np.array,
y_train: np.array,
x_val: np.array,
y_val: np.array,
maximum_epochs: int,
early_stop_epochs: int,
deeper_model: bool,
*args,
**kwargs
):
'''
Create ResNet-inspired CNN Model
'''
init_kernel_size = kwargs.get('init_kernel_size')
checkpoint_path = kwargs.get('checkpoint_path')
save_best_only = kwargs.get('save_best_only')
if deeper_model == True:
model = self.create_model(x_train, y_train, init_kernel_size)
else:
model = self.create_simpler_model(x_train, y_train, init_kernel_size)
# To use trained weights
if checkpoint_path != None:
model.load_weights(checkpoint_path)
if x_val is not None and y_val is not None:
# Naming convention to use for saved weights
checkpoint_name = 'Weights-{epoch:03d}--{loss:.2f}--{val_loss:.2f}.hdf5'
# Define callbacks
callbacks_list = [
EarlyStopping(
monitor='val_loss',
patience=early_stop_epochs
),
ModelCheckpoint(
filepath=checkpoint_name,
monitor='val_loss',
save_best_only=save_best_only
)
]
# Fit the model
blackbox = model.fit(
x=x_train,
y=y_train,
batch_size=32,
epochs=maximum_epochs,
validation_data=(x_val, y_val),
shuffle=True,
verbose=1,
callbacks=callbacks_list
)
# Otherwise training the model with full training dataset
elif x_val is None and y_val is None:
# Naming convention to use for saved weights
checkpoint_name = 'Weights-{epoch:03d}--{loss:.2f}.hdf5'
# Define callbacks
callbacks_list = [
EarlyStopping(
monitor='loss',
patience=early_stop_epochs
),
ModelCheckpoint(
filepath=checkpoint_name,
monitor='loss',
save_best_only=save_best_only
)
]
# Fit the model
blackbox = model.fit(
x=x_train,
y=y_train,
batch_size=32,
epochs=maximum_epochs,
shuffle=True,
verbose=1,
callbacks=callbacks_list
)
return model