-
Notifications
You must be signed in to change notification settings - Fork 5
/
nn.py
335 lines (285 loc) · 15.5 KB
/
nn.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
import tensorflow as tf
from keras.models import Model, Sequential
from keras.layers import Layer, Input, Reshape, Dense, Conv1D, Convolution1D, Dropout
from keras.layers import Activation, AveragePooling1D, MaxPooling1D, Flatten, Lambda, Multiply, Add
from keras.regularizers import l1, l2
from keras import backend as K
class CompositionVAE:
"""
Variational autoencoder to dense chemical compositions
"""
def __init__(
self, original_dim=200, latent_dim=64, initializer='he_normal',
last_layer_activation='linear', activation='relu', optimizer='rmsprop',
num_hidden=[1024, 512], dropout_rate=0.02, l1=1e-5, l2=1e-5,
epsilon_std=1e-4,
):
self.original_dim = original_dim
self._input_shape = (self.original_dim,)
self.latent_dim = latent_dim
self.initializer = initializer
self.last_layer_activation = last_layer_activation
self.activation = activation
self.num_hidden = num_hidden
self.dropout_rate = dropout_rate
self.epsilon_std = epsilon_std
self.l1 = l1
self.l2 = l2
self.optimizer = optimizer
self.vae, self.encoder, self.decoder = self.compile()
def compile(self):
input_tensor = Input(shape=self._input_shape, name='input')
layer_1 = Dense(units=self.num_hidden[0], kernel_regularizer=l1(self.l1),
kernel_initializer=self.initializer, activation=self.activation)(input_tensor)
dropout_1 = Dropout(self.dropout_rate)(layer_1)
layer_2 = Dense(units=self.num_hidden[1], kernel_regularizer=l1(self.l1),
kernel_initializer=self.initializer, activation=self.activation)(dropout_1)
dropout_2 = Dropout(self.dropout_rate)(layer_2)
z_mu = Dense(units=self.latent_dim, activation="linear", name='mean_output',
kernel_initializer=self.initializer,
kernel_regularizer=l1(self.l1))(dropout_2)
z_log_var = Dense(units=self.latent_dim, activation='linear', name='sigma_output',
kernel_initializer=self.initializer,
kernel_regularizer=l1(self.l1))(dropout_2)
z_mu, z_log_var = KLDivergenceLayer()([z_mu, z_log_var])
z_sigma = Lambda(lambda t: tf.exp(.5 * t))(z_log_var)
eps = Input(
tensor=tf.random_normal(mean=0., stddev=self.epsilon_std, shape=(tf.shape(z_mu)[0], self.latent_dim)))
z_eps = Multiply()([z_sigma, eps])
z = Add()([z_mu, z_eps])
decoder = Sequential(name='output')
decoder.add(Dense(units=self.num_hidden[1], kernel_regularizer=l1(self.l1),
kernel_initializer=self.initializer, activation=self.activation, input_dim=self.latent_dim))
decoder.add(Dropout(self.dropout_rate))
decoder.add(Dense(units=self.num_hidden[0], kernel_regularizer=l2(self.l2),
kernel_initializer=self.initializer, activation=self.activation))
decoder.add(Dropout(self.dropout_rate))
decoder.add(Dense(units=self.original_dim, kernel_regularizer=l2(self.l2),
kernel_initializer=self.initializer, activation=self.last_layer_activation, name='output'))
x_pred = decoder(z)
x_pred = Reshape((self.original_dim,), input_shape=(self.original_dim, 1))(x_pred)
vae = Model(inputs=[input_tensor, eps], outputs=x_pred)
vae.compile(optimizer=self.optimizer, loss=nll)
encoder = Model(input_tensor, z_mu)
print(vae.summary())
return vae, encoder, decoder
class XRDVAE:
"""
Convolutional variational autoencoder to dense 1D diffraction patterns
Original implementation: https://github.com/henrysky/astroNN (arXiv:1808.04428)
"""
def __init__(
self, original_dim=200, latent_dim=64, initializer='he_normal',
filter_len=2, last_layer_activation='linear', activation='relu',
num_filters=[16, 32, 64], num_hidden=[1024, 512], dropout_rate=0.02,
batch_size=64, epsilon_std=1e-4, l1=1e-5, l2=1e-5, pool_length=3,
optimizer='rmsprop',
):
self.original_dim = original_dim
self._input_shape = (self.original_dim,)
self.latent_dim = latent_dim
self.initializer = initializer
self.filter_len = filter_len
self.last_layer_activation = last_layer_activation
self.activation = activation
self.num_filters = num_filters
self.num_hidden = num_hidden
self.dropout_rate = dropout_rate
self.batch_size = batch_size
self.epsilon_std = epsilon_std
self.l1 = l1
self.l2 = l2
self.pool_length = pool_length
self.optimizer = optimizer
self.vae, self.encoder, self.decoder = self.compile()
def compile(self):
input_tensor = Input(shape=self._input_shape, name='input')
input_internal = Reshape((self.original_dim, 1), input_shape=self._input_shape)(input_tensor)
cnn_layer_1 = Conv1D(kernel_initializer=self.initializer, activation=self.activation, padding="same",
filters=self.num_filters[0],
kernel_size=self.filter_len, kernel_regularizer=l2(self.l2))(input_internal)
dropout_1 = Dropout(self.dropout_rate)(cnn_layer_1)
maxpool_1 = MaxPooling1D(pool_size=self.pool_length)(dropout_1)
cnn_layer_2 = Conv1D(kernel_initializer=self.initializer, activation=self.activation, padding="same",
filters=self.num_filters[1],
kernel_size=self.filter_len, kernel_regularizer=l2(self.l2))(maxpool_1)
dropout_2 = Dropout(self.dropout_rate)(cnn_layer_2)
maxpool_2 = MaxPooling1D(pool_size=self.pool_length)(dropout_2)
cnn_layer_3 = Conv1D(kernel_initializer=self.initializer, activation=self.activation, padding="same",
filters=self.num_filters[2],
kernel_size=self.filter_len, kernel_regularizer=l2(self.l2))(maxpool_2)
dropout_3 = Dropout(self.dropout_rate)(cnn_layer_3)
maxpool_3 = MaxPooling1D(pool_size=self.pool_length)(dropout_3)
flattener = Flatten()(maxpool_3)
layer_4 = Dense(units=self.num_hidden[0], kernel_regularizer=l1(self.l1),
kernel_initializer=self.initializer, activation=self.activation)(flattener)
dropout_3 = Dropout(self.dropout_rate)(layer_4)
layer_5 = Dense(units=self.num_hidden[1], kernel_regularizer=l1(self.l1),
kernel_initializer=self.initializer, activation=self.activation)(dropout_3)
dropout_4 = Dropout(self.dropout_rate)(layer_5)
z_mu = Dense(units=self.latent_dim, activation="linear", name='mean_output',
kernel_initializer=self.initializer,
kernel_regularizer=l1(self.l1))(dropout_4)
z_log_var = Dense(units=self.latent_dim, activation='linear', name='sigma_output',
kernel_initializer=self.initializer,
kernel_regularizer=l1(self.l1))(dropout_4)
z_mu, z_log_var = KLDivergenceLayer()([z_mu, z_log_var])
z_sigma = Lambda(lambda t: tf.exp(.5 * t))(z_log_var)
eps = Input(
tensor=tf.random_normal(mean=0., stddev=self.epsilon_std, shape=(tf.shape(z_mu)[0], self.latent_dim)))
z_eps = Multiply()([z_sigma, eps])
z = Add()([z_mu, z_eps])
decoder = Sequential(name='output')
decoder.add(Dense(units=self.num_hidden[1], kernel_regularizer=l1(self.l1),
kernel_initializer=self.initializer, activation=self.activation, input_dim=self.latent_dim))
decoder.add(Dropout(self.dropout_rate))
decoder.add(Dense(units=self._input_shape[0]*self.num_filters[1], kernel_regularizer=l2(self.l2),
kernel_initializer=self.initializer, activation=self.activation))
decoder.add(Dropout(self.dropout_rate))
output_shape = (self.batch_size, self._input_shape[0], self.num_filters[1])
decoder.add(Reshape(output_shape[1:]))
decoder.add(Conv1D(kernel_initializer=self.initializer, activation=self.activation, padding="same",
filters=self.num_filters[2],
kernel_size=self.filter_len, kernel_regularizer=l2(self.l2)))
decoder.add(Dropout(self.dropout_rate))
decoder.add(Conv1D(kernel_initializer=self.initializer, activation=self.activation, padding="same",
filters=self.num_filters[1],
kernel_size=self.filter_len, kernel_regularizer=l2(self.l2)))
decoder.add(Dropout(self.dropout_rate))
decoder.add(Conv1D(kernel_initializer=self.initializer, activation=self.activation, padding="same",
filters=self.num_filters[0],
kernel_size=self.filter_len, kernel_regularizer=l2(self.l2)))
decoder.add(Conv1D(kernel_initializer=self.initializer, activation=self.last_layer_activation, padding="same",
filters=1, kernel_size=self.filter_len, name='output'))
x_pred = decoder(z)
x_pred = Reshape((self.original_dim,), input_shape=(self.original_dim, 1))(x_pred)
vae = Model(inputs=[input_tensor, eps], outputs=x_pred)
vae.compile(optimizer=self.optimizer, loss=nll)
encoder = Model(input_tensor, z_mu)
print(vae.summary())
return vae, encoder, decoder
class SpaceGroupClassifier():
"""
Fully-connected neural network to classify 1D diffraction patterns
Original implementation: https://doi.org//10.1107/S205225251700714X/fc5018sup1.pdf
Park, Woon Bae and Chung, Jiyong and Jung, Jaeyoung and Sohn, Keemin and Singh, Satendra Pal and Pyo, Myoungho and Shin, Namsoo and Sohn, Kee-Sun
Classification of crystal structure using a convolutional neural network
IUCrJ, 4, 4, 2017, 486--494, 10.1107/S205225251700714X
"""
def __init__(
self, input_shape=(8001, 1),
num_filters=[80, 80, 80], strides=[5, 5, 2], kernel_size=[100, 50, 25],
num_hidden=[2300, 1150], conv_dropout=0.3, conn_dropout=0.5, pool_size=3,
activation='relu', last_layer_activation='softmax', optimizer='Adam',
initializer='he_normal'
):
self.input_shape = input_shape
self.num_filters = num_filters
self.strides = strides
self.kernel_size = kernel_size
self.conv_dropout = conv_dropout
self.conn_dropout = conn_dropout
self.num_hidden = num_hidden
self.activation = activation
self.last_layer_activation = last_layer_activation
self.pool_size = pool_size
self.optimizer = optimizer
self.initializer = initializer
self.model = self.compile()
def compile(self):
model = Sequential()
model.add(Convolution1D(self.num_filters[0], self.kernel_size[0], strides=self.strides[0], padding='same',
input_shape=self.input_shape, kernel_initializer=self.initializer))
model.add(Activation(self.activation))
model.add(Dropout(self.conv_dropout))
model.add(AveragePooling1D(pool_size=self.pool_size, strides=None))
model.add(Convolution1D(self.num_filters[1], self.kernel_size[1], strides=self.strides[1], padding='same',
kernel_initializer=self.initializer))
model.add(Activation(self.activation))
model.add(Dropout(self.conv_dropout))
model.add(AveragePooling1D(pool_size=self.pool_size, strides=None))
model.add(Convolution1D(self.num_filters[2], self.kernel_size[2], strides=self.strides[2], padding='same',
kernel_initializer=self.initializer))
model.add(Activation(self.activation))
model.add(Dropout(self.conv_dropout))
model.add(AveragePooling1D(pool_size=self.pool_size, strides=None))
model.add(Flatten())
model.add(Dense(self.num_hidden[0], kernel_initializer=self.initializer))
model.add(Activation(self.activation))
model.add(Dropout(self.conn_dropout))
model.add(Dense(self.num_hidden[1], kernel_initializer=self.initializer))
model.add(Activation(self.activation))
model.add(Dropout(self.conn_dropout))
model.add(Dense(230))
model.add(Activation(self.last_layer_activation))
model.compile(loss='categorical_crossentropy', optimizer=self.optimizer, metrics=['accuracy'])
print(model.summary())
return model
class FormationEnergyEstimator:
"""
Fully-connected neural network to estimate stability of structure (formation energy )
"""
def __init__(
self, input_dim=88, kernel_initializer='he_normal',
last_layer_activation='sigmoid', activation='relu',
num_hidden=64, dropout_rate=0.5, optimizer='rmsprop'
):
self.input_dim = input_dim
self.last_layer_activation=last_layer_activation
self.kernel_initializer=kernel_initializer
self.activation=activation
self.num_hidden = num_hidden
self.dropout_rate = dropout_rate
self.optimizer = optimizer
self.model = self.compile()
def compile(self):
model = Sequential()
model.add(Dense(self.num_hidden, input_dim=self.input_dim, activation=self.activation,
kernel_initializer=self.kernel_initializer))
model.add(Dropout(self.dropout_rate))
model.add(Dense(self.num_hidden, activation=self.activation, kernel_initializer=self.kernel_initializer))
model.add(Dropout(self.dropout_rate))
model.add(Dense(self.num_hidden, activation=self.activation, kernel_initializer=self.kernel_initializer))
model.add(Dropout(self.dropout_rate))
model.add(Dense(1, activation=self.last_layer_activation, kernel_initializer=self.kernel_initializer))
model.compile(loss='binary_crossentropy', optimizer=self.optimizer, metrics=['accuracy'])
print(model.summary())
return model
class KLDivergenceLayer(Layer):
"""
| Identity transform layer that adds KL divergence to the final model losses.
| KL divergence used to force the latent space match the prior (in this case its unit gaussian)
:return: A layer
:rtype: object
:History: 2018-Feb-05 - Written - Henry Leung (University of Toronto)
"""
def __init__(self, name=None, **kwargs):
self.is_placeholder = True
if not name:
prefix = self.__class__.__name__
name = prefix + '_' + str(K.get_uid(prefix))
super().__init__(name=name, **kwargs)
def call(self, inputs, training=None):
"""
:Note: Equivalent to __call__()
:param inputs: Tensor to be applied, concatenated tf.tensor of mean and std in latent space
:type inputs: tf.Tensor
:return: Tensor after applying the layer
:rtype: tf.Tensor
"""
mu, log_var = inputs
kl_batch = - .5 * tf.reduce_sum(1 + log_var - tf.square(mu) - tf.exp(log_var), axis=-1)
self.add_loss(tf.reduce_mean(kl_batch), inputs=inputs)
return inputs
def get_config(self):
"""
:return: Dictionary of configuration
:rtype: dict
"""
config = {'None': None}
base_config = super().get_config()
return {**dict(base_config.items()), **config}
def compute_output_shape(self, input_shape):
return input_shape
def nll(y_true, y_pred):
return K.sum(K.binary_crossentropy(y_true, y_pred), axis=-1)