Generate_Algorithm/avb.py at master · ssliar/Generate_Algorithm · GitHub

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
'''This script demonstrates how to build a adversarial variational bayes with Keras.
'''
import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm

from keras.layers import Input, Dense, Lambda, Activation, Concatenate, Dot
from keras.models import Model
from keras.regularizers import l2
from keras import backend as K
from keras import objectives
from keras.datasets import mnist
from keras.optimizers import *

np.random.seed(1111)  # for reproducibility

training = True

batch_size = 50
n = 784 # for datapoint
m = 2 # for hidden variables
l = 5 # for random noise
hidden_dim = 256
epochs = 50
epsilon_std = 1.0
loss = 'categorical_crossentorpy' # 'mse' or 'categorical_crossentropy'

decay = 1e-4 # weight decay, a.k. l2 regularization
use_bias = True

## Encoder
x = Input(shape=(n,))
g = Input(shape=(l,))
x_g = Concatenate(-1)([x, g])
h_encoded = Dense(hidden_dim, kernel_regularizer=l2(decay), bias_regularizer=l2(decay), use_bias=use_bias, activation='relu')(x_g)
z = Dense(m, activation='relu', kernel_regularizer=l2(decay), bias_regularizer=l2(decay), use_bias=use_bias)(h_encoded)

encoder = Model([x, g], z)

## Decoder
decoder_h = Dense(hidden_dim, kernel_regularizer=l2(decay), bias_regularizer=l2(decay), use_bias=use_bias, activation='relu')
decoder_mean = Dense(n, kernel_regularizer=l2(decay), bias_regularizer=l2(decay), use_bias=use_bias, activation='sigmoid')
h_decoded = decoder_h(z)
x_hat = decoder_mean(h_decoded)

rec = Model([x, g], x_hat)
rec.compile(optimizer=Adam(1e-3), loss='binary_crossentropy')

## Discriminator
x_h = Dense(hidden_dim, kernel_regularizer=l2(decay), bias_regularizer=l2(decay), use_bias=use_bias, activation='relu')
z_h = Dense(hidden_dim, kernel_regularizer=l2(decay), bias_regularizer=l2(decay), use_bias=use_bias, activation='relu')

### gan_g
x_h.trainable = False
z_h.trainable = False
T = Dot(-1)([x_h(x), z_h(z)])
gan_g = Model([x, g], T)
gan_g.compile(optimizer=Adam(1e-4), loss=lambda y_true, y_predict: -K.mean(y_predict, -1))

### gan_d
x_h.trainable = True
z_h.trainable = True
x = Input(shape=(n,))
z = Input(shape=(m,))
fake = Dot(-1)([x_h(x), z_h(z)])
fake = Activation('sigmoid')(fake)
gan_d = Model([x, z], fake)
gan_d.compile(optimizer=Adam(1e-4), loss='binary_crossentropy')

if training:
    # train the VAE on MNIST digits
    (x_train, y_train), (x_test, y_test) = mnist.load_data()

    x_train = x_train.astype('float32') / 255
    x_test = x_test.astype('float32') / 255
    x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:])))
    x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:])))

    ids = range(len(x_train))
    for i in range(epochs):
        print('Epoch {}:'.format(i + 1))
        rec_loss = []
        gen_loss = []
        dis_loss = []
        for s in range(0, len(ids), batch_size):
            x = x_train[ids[s:s+batch_size]]
            y = y_train[ids[s:s+batch_size]]
            bs = len(x)

            eps = np.random.randn(bs, l)
            z = np.random.randn(bs, m)

            # reconstruction
            loss = rec.train_on_batch([x, eps], x)
            rec_loss.append(loss)

            # encoder
            loss = gan_g.train_on_batch([x, eps], [1] * bs)
            gen_loss.append(loss)

            # discriminator
            z_fake = encoder.predict([x, eps])
            x = np.concatenate([x, x], axis=0)
            z = np.concatenate([z, z_fake], axis=0)
            y = np.asarray([1] * bs + [0] * bs, dtype='float32')

            loss = gan_d.train_on_batch([x, z], y)
            dis_loss.append(loss)

            if s % 1000 == 0:
                print('rec loss: {}, gen loss: {}, dis loss: {}'.\
                       format(np.mean(rec_loss), np.mean(gen_loss), np.mean(dis_loss)))

        print('rec loss: {}, gen loss: {}, dis loss: {}'.\
               format(np.mean(rec_loss), np.mean(gen_loss), np.mean(dis_loss)))
        rec.save_weights('weights_{}.h5'.format(i))

# display a 2D manifold of the digits
n = 15  # figure with 15x15 digits
digit_size = 28
figure = np.zeros((digit_size * n, digit_size * n))
# linearly spaced coordinates on the unit square were transformed through the inverse CDF (ppf) of the Gaussian
# to produce values of the latent variables z, since the prior of the latent space is Gaussian
grid_x = norm.ppf(np.linspace(0.05, 0.95, n))
grid_y = norm.ppf(np.linspace(0.05, 0.95, n))

z = Input(shape=(m,))
h_decoded = decoder_h(z)
x_hat = decoder_mean(h_decoded)
decoder = Model(z, x_hat)
decoder.load_weights('weights_8.h5', by_name=True)

for i, yi in enumerate(grid_x):
    for j, xi in enumerate(grid_y):
        z_sample = np.array([[xi, yi]])
        x_decoded = decoder.predict(z_sample)
        digit = x_decoded[0].reshape(digit_size, digit_size)
        figure[i * digit_size: (i + 1) * digit_size,
               j * digit_size: (j + 1) * digit_size] = digit

fig = plt.figure(figsize=(10, 10))
plt.imshow(figure, cmap='Greys_r')
plt.show()
fig.savefig('x.png')