-
Notifications
You must be signed in to change notification settings - Fork 0
/
autoencoder.py
197 lines (154 loc) · 6.69 KB
/
autoencoder.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
# -*- coding: utf-8 -*-
import numpy as np
import os
from keras.layers import containers, AutoEncoder, Dense, Dropout
from keras import models
from keras.layers.convolutional import Convolution2D, MaxPooling2D, UpSampling2D
from keras.layers.core import Dense, Activation, Dropout, Flatten, Reshape
from autoencoder_layers import DependentDense, Deconvolution2D, DePool2D
def compress_data_by_autoencoder(X, target_dim, hidden_dim, weightsFile=None):
input_dim = X.shape[1]
# input shape: (nb_samples, 32)
encoder = containers.Sequential([
Dense(hidden_dim, input_dim=input_dim, activation='relu'),
Dropout(0.2),
Dense(hidden_dim, activation='relu'),
Dropout(0.2),
Dense(hidden_dim, activation='relu'),
Dropout(0.2),
Dense(target_dim, activation='relu'),
])
decoder = containers.Sequential([
Dense(hidden_dim, input_dim=target_dim, activation='relu'),
Dropout(0.2),
Dense(hidden_dim, activation='relu'),
Dropout(0.2),
Dense(hidden_dim, activation='relu'),
Dropout(0.2),
Dense(input_dim, activation='relu')
])
autoencoder = AutoEncoder(encoder=encoder, decoder=decoder, output_reconstruction=True)
model = models.Sequential()
model.add(autoencoder)
# training the autoencoder:
model.compile(optimizer='sgd', loss='mse')
if ( type(weightsFile) is str and os.path.isfile(weightsFile) ):
model.load_weights(weightsFile)
error = 1
max_nb_epoch = 1000
epoch = 0
while (error > 0.005 and epoch < max_nb_epoch):
fited = model.fit(X, X, nb_epoch=1, batch_size=64) #
error = fited.totals['loss']/64
epoch += 1
# predicting compressed representations of inputs:
autoencoder.output_reconstruction = False # the model has to be recompiled after modifying this property
model.compile(optimizer='sgd', loss='mse')
representations = model.predict(X)
if ( type(weightsFile) is str ):
model.save_weights(weightsFile, overwrite=True)
return representations
##########################################################################
def compress_data_by_conv_autoencoder(X, target_dim, hidden_dim, weightsFile=None):
input_dim = X.shape[1]
dataset_size = X.shape[0]
nb_filters = 5
nb_conv = 5
nb_pool = 2
X = X.reshape(dataset_size, 1, input_dim, 1)
hidden_dense_layer1 = Dense(hidden_dim)
hidden_dense_layer2 = Dense(hidden_dim)
conv_layer1 = Convolution2D(nb_filters, nb_conv, nb_conv,
border_mode='same',
input_shape=(1, input_dim, 1))
mp_layer1 = MaxPooling2D(pool_size=(nb_pool, 1))
# conv_layer2 = Convolution2D(nb_filters, nb_conv, 1,
# border_mode='same',
# input_shape=(1, input_dim, 1))
# conv_layer3 = Convolution2D(nb_filters, nb_conv, 1,
# border_mode='same',
# input_shape=(1, input_dim, 1))
#
# conv_layer4 = Convolution2D(nb_filters, nb_conv, 1,
# border_mode='same',
# input_shape=(1, input_dim, 1))
#mp_layer2 = MaxPooling2D(pool_size=(nb_pool, 1))
encoder = models.Sequential()
encoder.add(conv_layer1)
encoder.add(Activation("relu"))
# encoder.add(Dropout(0.2))
#
# encoder.add(conv_layer2)
# encoder.add(Activation("relu"))
encoder.add(mp_layer1)
#encoder.add(mp_layer2)
encoder.add(Dropout(0.2))
encoder.add(Flatten())
encoder.add(hidden_dense_layer1)
encoder.add(Activation("relu"))
encoder.add(Dropout(0.2))
encoder.add(hidden_dense_layer2)
encoder.add(Activation("relu"))
encoder.add(Dropout(0.2))
encoder.add(Dense(target_dim))
encoder.add(Activation("relu"))
######### decoder ###############
decoder = models.Sequential()
decoder.add(Dense(hidden_dim, input_dim=target_dim, activation='relu'))
decoder.add(Dropout(0.2))
decoder.add(DependentDense( (nb_filters * input_dim//nb_pool), hidden_dense_layer1))
decoder.add(Activation('relu'))
decoder.add(Dropout(0.2))
decoder.add(Reshape((nb_filters, input_dim//nb_pool, 1)))
decoder.add(DePool2D(mp_layer1, size=(nb_pool, 1)))
decoder.add(Deconvolution2D(conv_layer1, border_mode='same'))
decoder.add(Activation('relu'))
# decoder.add(Deconvolution2D(conv_layer4, border_mode='same'))
# decoder.add(Activation('relu'))
autoencoder = AutoEncoder(encoder=encoder, decoder=decoder, output_reconstruction=True)
model = models.Sequential()
model.add(autoencoder)
# training the autoencoder:
model.compile(optimizer='adam', loss='mse')
if ( type(weightsFile) is str and os.path.isfile(weightsFile) ):
model.load_weights(weightsFile)
# !!!!!!!!!!!!!!!!!!
model.fit(X, X, nb_epoch=150, batch_size=100) #
# !!!!!!!!!!!!!!!!!!
# predicting compressed representations of inputs:
autoencoder.output_reconstruction = False # the model has to be recompiled after modifying this property
model.compile(optimizer='sgd', loss='mse')
representations = model.predict(X)
if ( type(weightsFile) is str ):
model.save_weights(weightsFile, overwrite=True)
return representations
"""
dataset = { "input": datasetOfSpikes, "target": datasetOfSpikes }
layers = [
{"n": self.window_size*resamplingFactor, "bias": True, },
{"n": 40, "bias": True, },
{"n": 40, "bias": True, },
{"n": 10, "bias": True, },
{"n": 40, "bias": True,},
{"n": 40, "bias": True, },
{"n": self.window_size*resamplingFactor, "bias": True,}
]
autoencoder = mynet.Network(layers, dataset["input"].shape[1])
if ( loadWeights and os.path.exists(weightsFile) ):
weightsData = io.loadmat(weightsFile)
weights = []
for idx in range(len(layers)):
weights.append(weightsData[str(idx)])
autoencoder.setWeights(weights)
save_err = autoencoder.trainOnDataset(dataset, 0.001, 100000, lr, 100)
print (save_err)
weights = autoencoder.getWeights()
if (saveWeights):
weightsData = {"data":{}}
for idx, l in enumerate(weights):
weightsData["data"][str(idx)] = l
io.savemat(weightsFile, weightsData["data"])
coder = mynet.Network(layers[0:4], dataset["input"].shape[1])
coder.setWeights(weights[0:4])
compress_data = coder.activate(dataset["input"])
"""