/
model.py
173 lines (141 loc) · 6.45 KB
/
model.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
# import pandas as pd
import numpy as np
from joblib import load as jLoad, dump# , Parallel, delayed
# from matplotlib import pyplot as plt
from keras.models import Sequential
from keras.layers import Dense, Activation, Flatten, Reshape, Conv2D,\
Conv2DTranspose, UpSampling2D, LeakyReLU, Dropout,\
BatchNormalization, SeparableConv2D
from keras.optimizers import RMSprop# , Adam
class Model:
def __init__(self, img_rows, img_cols, channel=1, discriminator=None,\
generator=None, adversarialModel=None,\
discriminatorModel=None):
self.img_rows = img_rows
self.img_cols = img_cols
self.channel = channel
self.__discriminator__ = discriminator
self.__generator__ = generator
self.__am__ = adversarialModel
self.__dm__ = discriminatorModel
def discriminator(self):
if not self.__discriminator__:
dropout = 0.2
convOut = 32
input_shape = (self.img_rows, self.img_cols, self.channel)
self.__discriminator__ = Sequential()
self.__discriminator__.add(Conv2D(convOut, 5, strides=(1, 1),\
input_shape=input_shape, padding='same'))
self.__discriminator__.add(LeakyReLU(alpha=0.12))
self.__discriminator__.add(Dropout(dropout))
self.__discriminator__.add(Conv2D(convOut * 2, 5, strides=(1, 1),\
padding='valid'))
self.__discriminator__.add(LeakyReLU(alpha=0.2))
self.__discriminator__.add(Dropout(dropout * 1.4))
self.__discriminator__.add(SeparableConv2D(convOut * 4, 5,\
strides=(1, 1)))
self.__discriminator__.add(LeakyReLU(alpha=0.2))
self.__discriminator__.add(Dropout(dropout * 1.6))
self.__discriminator__.add(Conv2D(convOut * 8, 5, strides=(1, 1),\
padding='valid'))
self.__discriminator__.add(LeakyReLU(alpha=0.15))
self.__discriminator__.add(Dropout(dropout / 1.3))
self.__discriminator__.add(Flatten())
self.__discriminator__.add(Dense(10))
self.__discriminator__.add(Activation('relu'))
self.__discriminator__.add(Dense(1))
self.__discriminator__.add(Activation('sigmoid'))
self.__discriminator__.summary()
return self.__discriminator__
def generator(self):
if not self.__generator__:
dropout = 0.2
depth = 4 * 32
self.__generator__ = Sequential()
self.__generator__.add(Dense(62*50*depth, input_dim=100))
self.__generator__.add(BatchNormalization(momentum=0.9))
self.__generator__.add(Activation('relu'))
self.__generator__.add(Reshape((62, 50, depth)))
self.__generator__.add(Dropout(dropout))
self.__generator__.add(UpSampling2D())
self.__generator__.add(Conv2DTranspose(int(depth/2), 5,\
padding='same'))
self.__generator__.add(BatchNormalization(momentum=0.9))
self.__generator__.add(Activation('relu'))
self.__generator__.add(UpSampling2D())
self.__generator__.add(Conv2DTranspose(int(depth/4), 5,\
padding='same'))
self.__generator__.add(BatchNormalization(momentum=0.9))
self.__generator__.add(Activation('relu'))
self.__generator__.add(Conv2DTranspose(int(depth/8), 5,\
padding='same'))
self.__generator__.add(BatchNormalization(momentum=0.9))
self.__generator__.add(Activation('relu'))
self.__generator__.add(Conv2DTranspose(1, 5, padding='same'))
self.__generator__.add(Activation('sigmoid'))
self.__generator__.summary()
return self.__generator__
def adversarialModel(self):
if not self.__am__:
optimizer = RMSprop(lr=1e-5, decay=3e-8)
self.__am__ = Sequential()
self.__am__.add(self.generator())
self.__am__.add(self.discriminator())
self.__am__.compile(optimizer=optimizer, loss='binary_crossentropy',\
metrics=['accuracy'])
return self.__am__
def discriminatorModel(self):
if not self.__dm__:
optimizer = RMSprop(lr=0.0002, decay=6e-8)
self.__dm__ = Sequential()
self.__dm__.add(self.discriminator())
self.__dm__.compile(loss='binary_crossentropy', optimizer=optimizer,\
metrics=['accuracy'])
return self.__dm__
def save(path, model):
print('Saving to %s...'%path)
dump(model, path)
def load(path=None, img_rows=248, img_cols=200):
if not path:
return Model(img_rows, img_cols)
print('Loading model from %s...'%path)
return jLoad(path)
def train(X, _y, trainSteps=1000, batchSize=2):
X_train = X.values.reshape((-1, 250, 200))[:, :248, :]
# y_train = _y.values.reshape((-1, 250, 200))[:, :248, :]
model = load()
am = model.adversarialModel()
dm = model.discriminatorModel()
gen = model.generator()
def step(iteration):
images_train = X_train[np.random.randint(0, X_train.shape[0],\
size=batchSize), :, :].reshape((batchSize, 248, 200, -1))
print('shape of images_train', images_train.shape)
noise = np.random.uniform(-1.0, 1.0, size=[batchSize, 100])
print('shape of noise', noise.shape)
images_fake = gen.predict(noise)# .reshape((batchSize, 248, 200))
print('shape of images_fake', images_fake.shape)
x = np.concatenate((images_train, images_fake))
print('images_train & images_fake concatenated')
y = np.ones([2*batchSize, 1])
y[batchSize:, :] = 0
print('y is', y)
print('Training on y')
d_loss = dm.train_on_batch(x, y)
y = np.ones([batchSize, 1])
noise = np.random.uniform(-1.0, 1.0, size=[batchSize, 100])
a_loss = am.train_on_batch(noise, y)
log_mesg = "%d: [D loss: %f, acc: %f]" % (iteration, d_loss[0], d_loss[1])
log_mesg = "%s [A loss: %f, acc: %f]" % (log_mesg, a_loss[0], a_loss[1])
print(log_mesg)
for i in range(trainSteps):
step(i)
dump(model, 'Lopez.model')
def test(dataset):
print('Testing dataset')
# 1.6e6/4875
'''
62.5 * 50 * 128 * 4 = 1.6e6 = 62 * 50 * 4 * 128 + .5 * 50 * 4 * 128
==> 62.5 * 50 * 128 * 4 = 1.6e6 = 62 * 50 * 4 * 129.0322580645
62 * 50 * 4 * 129.03225806451611609950000000000004999...
'''