-
Notifications
You must be signed in to change notification settings - Fork 0
/
vae.py
254 lines (194 loc) · 7.62 KB
/
vae.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
import numpy as np
import tensorflow as tf
import os
from tqdm import tqdm
import time
import matplotlib.pyplot as plt
class VAE(tf.keras.Model):
def __init__(self, latent_dim):
super().__init__()
self.latent_dim = latent_dim
self.encoder = tf.keras.Sequential(
[
tf.keras.layers.InputLayer(input_shape=(FRAME_SIZE, FRAME_SIZE, 3)),
tf.keras.layers.Conv2D(
filters=32, kernel_size=4, strides=2, activation='relu'), # 31x31
tf.keras.layers.BatchNormalization(),
tf.keras.layers.Conv2D(
filters=64, kernel_size=4, strides=2, activation='relu'), # 14x14
tf.keras.layers.BatchNormalization(),
tf.keras.layers.Conv2D(
filters=128, kernel_size=4, strides=2, activation='relu'), # 6x6
tf.keras.layers.BatchNormalization(),
tf.keras.layers.Conv2D(
filters=256, kernel_size=4, strides=2, activation='relu'), # 2x2
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(latent_dim + latent_dim),
],
name = 'encoder'
)
self.decoder = tf.keras.Sequential(
[
tf.keras.layers.InputLayer(input_shape=(latent_dim,)),
tf.keras.layers.Dense(units=2**2*256, activation=tf.nn.relu),
tf.keras.layers.Reshape(target_shape=(-1, 1, 2**2*256)),
tf.keras.layers.Conv2DTranspose(
filters=128,
kernel_size=5,
strides=2,
activation='relu'),
tf.keras.layers.Conv2DTranspose(
filters=64,
kernel_size=5,
strides=2,
activation='relu'),
tf.keras.layers.Conv2DTranspose(
filters=32,
kernel_size=6,
strides=2,
activation='relu'),
tf.keras.layers.Conv2DTranspose(
filters=3,
kernel_size=6,
strides=2,
activation='sigmoid'),
],
name = 'decoder'
)
def call(self, x, training=True):
mu, logvar = self.encode(x, training)
z = self.reparameterize(mu, logvar)
return self.decode(z)
def summary(self):
self.encoder.summary()
self.decoder.summary()
@tf.function
def sample(self, eps=None):
if eps is None:
eps = tf.random.normal(shape=(100, self.latent_dim))
return self.decode(eps, apply_sigmoid=True)
def encode(self, x, training=True):
mu, logvar = tf.split(self.encoder(x, training=training), num_or_size_splits=2, axis=1)
return mu, logvar
def reparameterize(self, mu, logvar):
eps = tf.random.normal(shape=mu.shape)
return eps * tf.exp(logvar * .5) + mu
def decode(self, z, apply_sigmoid=False):
logits = self.decoder(z)
return logits # not logits
def latent(self, x):
mu, logvar = self.encode(x, training=False)
return self.reparameterize(mu, logvar)
def test(self, x):
n = 5
samples = np.random.choice(len(x), n**2, replace=False)
samples.sort()
x = x[samples,:]
preds = self.call(x, training=False)
preds = np.array(preds)
#print(preds)
imd = FRAME_SIZE
canvas_orig = np.empty((imd*n , 2*imd * n+1, 3))
for i in range(n):
batch_x = x[i*n:i*n+n]
g = preds[i*n:i*n+n]
for j in range(n):
canvas_orig[i * imd:(i + 1) * imd, j * imd:(j + 1) * imd] = \
batch_x[j].reshape([imd, imd, 3])
canvas_orig[i * imd :(i + 1) * imd, j * imd + n*imd+1:(j + 1) * imd + n*imd+1] = \
g[j].reshape([imd, imd, 3])
canvas_orig[:, n*imd:n*imd+1] = 1
print("Original Images")
plt.figure(figsize=(n*2+1, n))
plt.imshow(canvas_orig, origin="upper")
plt.draw()
plt.show()
def create_dataset(filelist, N=100, M=10000, T=1000): # N is 10000 episodes, M is number of timesteps
test_data = np.zeros((T, FRAME_SIZE, FRAME_SIZE, 3), dtype=np.uint8)
filename = filelist[-1]
raw_data = np.load(os.path.join(DATA_DIR, filename))['obs']
if T > len(raw_data):
print('T too large, check test dataset creation function')
T = len(raw_data)
test_data = raw_data[:T]
data = np.zeros((M*N, FRAME_SIZE, FRAME_SIZE, 3), dtype=np.uint8)
idx = 0
for i in range(N):
filename = filelist[i]
raw_data = np.load(os.path.join(DATA_DIR, filename))['obs']
l = len(raw_data)
if (idx+l) > (M*N):
data = data[:idx]
print('premature break')
break
l = min(l, M)
data[idx:idx+l] = raw_data[:l]
idx += l
if ((i+1) % 100 == 0):
print("loading file", i+1)
data = data[:idx]
return data, test_data
@tf.function
def compute_loss(model, x, beta):
mean, logvar = model.encode(x)
z = model.reparameterize(mean, logvar)
y = model.decode(z)
#rec_loss = tf.nn.sigmoid_cross_entropy_with_logits(logits=y, labels=x) # MSE/L2?
#rec_loss = -tf.math.reduce_sum(rec_loss) # , axis=[1, 2, 3])
rec_loss = tf.reduce_sum(tf.math.square(x - y))
rec_loss = tf.reduce_mean(rec_loss)
kl_loss = -0.5*tf.math.reduce_sum((1 + logvar - tf.square(mean) - tf.exp(logvar)))
# https://openreview.net/forum?id=Sy2fzU9gl
kl_loss *= beta[0]
kl_loss = tf.reduce_mean(kl_loss)
return tf.reduce_mean(kl_loss+rec_loss)
@tf.function
def compute_apply_gradients(model, x, optimizer, beta):
with tf.GradientTape() as tape:
loss = compute_loss(model, x, beta)
gradients = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(gradients, model.trainable_variables))
FRAME_SIZE = 64
if __name__=='__main__':
#model = VAE(32)
#model.summary()
#raise
print(tf.__version__)
NUM_EPOCH = 10
DATA_DIR = "record_car_racing"
model_save_path = "vae_ckpt"
if not os.path.exists(model_save_path):
os.makedirs(model_save_path)
batch_size = 128
filelist = os.listdir(DATA_DIR)
filelist.sort()
filelist = filelist #[0:1000]
dataset, test_data = create_dataset(filelist, N=5)
total_length = len(dataset)
print(total_length)
num_batches = int(np.floor(total_length/batch_size))
print("num_batches", num_batches)
optimizer = tf.keras.optimizers.Adam()
model = VAE(32)
model.summary()
test_data = test_data.astype(np.float32)
test_data /= 255
model.test(test_data)
epochs = 10
kl_warm_up_epochs = 1
for epoch in range(1, epochs + 1):
beta = np.array([1], dtype=np.float32)
print(epoch)
st = time.time()
for batch in tqdm(range(num_batches)):
if epoch <= kl_warm_up_epochs:
# KL Warm Up
# https://arxiv.org/abs/1602.02282
beta = np.array([(batch + (epoch-1)*num_batches) / (num_batches*kl_warm_up_epochs)], dtype=np.float32)
train_x = dataset[batch*batch_size:(batch+1)*batch_size]
train_x = train_x.astype(np.float32)
train_x /= 255
compute_apply_gradients(model, train_x, optimizer, beta)
model.test(test_data)
model.save_weights(os.path.join(model_save_path,'VAE_Epoch{epoch:04d}'.format(epoch=epoch)))
print('T:', time.time()-st)