-
Notifications
You must be signed in to change notification settings - Fork 0
/
run.py
184 lines (151 loc) · 6.06 KB
/
run.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
from __future__ import print_function
import tensorflow as tf
import numpy as np
import argparse
import time
import os
from model import autoencoder, decoder, encoder
from accumulator import Accumulator
from mnist import mnist_1000
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches
from sklearn.manifold import TSNE
#from scipy.misc import imsave
#from scipy.misc import imresize
parser = argparse.ArgumentParser()
parser.add_argument('--batch_size', type=int, default=100)
parser.add_argument('--n_epochs', type=int, default=100)
parser.add_argument('--savedir', type=str, default=None)
parser.add_argument('--mnist_path', type=str, default='./data')
parser.add_argument('--mode', type=str, default='train')
parser.add_argument('--gpu_num', type=int, default=0)
parser.add_argument('--zdim', type=int, default=2)
parser.add_argument("--scatter", default=False, action="store_true")
parser.add_argument("--reconstruct", default=False, action="store_true")
parser.add_argument("--manifold", default=False, action="store_true")
args = parser.parse_args()
os.environ['TF_ENABLE_WINOGRAD_NONFUSED'] = '1'
os.environ['CUDA_VISIBLE_DEVICES'] = str(args.gpu_num)
savedir = './results/run' \
if args.savedir is None else args.savedir
if not os.path.isdir(savedir):
os.makedirs(savedir)
# get data
xtr, ytr, xte, yte = mnist_1000(args.mnist_path)
# placeholders
x = tf.placeholder(tf.float32, [None, 784])
n_train_batches = int(1000/args.batch_size)
n_test_batches = int(1000/args.batch_size)
# models
net = autoencoder(x, args.zdim, True) # train
tnet = autoencoder(x, args.zdim, False, reuse=True) # test
# for visualization
z = tf.placeholder(tf.float32, [None, args.zdim])
tennet = encoder(x, args.zdim, reuse=True) # test encoder
tdenet = decoder(z, reuse=True) # test decoder
def train():
loss = -net['elbo'] # negative ELBO
global_step = tf.train.get_or_create_global_step()
lr = tf.train.piecewise_constant(tf.cast(global_step, tf.int32),
[int(n_train_batches*args.n_epochs/2)], [1e-3, 1e-4])
train_op = tf.train.AdamOptimizer(lr).minimize(loss,
global_step=global_step)
saver = tf.train.Saver(net['weights'])
logfile = open(os.path.join(savedir, 'train.log'), 'w')
sess = tf.Session()
sess.run(tf.global_variables_initializer())
# to run
train_logger = Accumulator('elbo')
train_to_run = [train_op, net['elbo']]
for i in range(args.n_epochs):
# shuffle the training data
idx = np.random.choice(range(1000), size=1000, replace=False)
xtr_ = xtr[idx]
# run the epoch
line = 'Epoch %d start, learning rate %f' % (i+1, sess.run(lr))
print('\n' + line)
logfile.write('\n' + line + '\n')
train_logger.clear()
start = time.time()
for j in range(n_train_batches):
bx = xtr_[j*args.batch_size:(j+1)*args.batch_size,:]
train_logger.accum(sess.run(train_to_run, {x:bx}))
train_logger.print_(header='train', epoch=i+1,
time=time.time()-start, logfile=logfile)
# save the model
logfile.close()
saver.save(sess, os.path.join(savedir, 'model'))
def test():
sess = tf.Session()
saver = tf.train.Saver(tnet['weights'])
saver.restore(sess, os.path.join(savedir, 'model'))
logger = Accumulator('elbo')
for j in range(n_test_batches):
bx = xte[j*args.batch_size:(j+1)*args.batch_size,:]
logger.accum(sess.run(tnet['elbo'], {x:bx}))
print()
logger.print_(header='test')
print()
def visualize():
sess = tf.Session()
saver = tf.train.Saver(tnet['weights'])
saver.restore(sess, os.path.join(savedir, 'model'))
visdir = os.path.join(savedir, 'vis')
if not os.path.isdir(visdir):
os.makedirs(visdir)
def _merge(images, size):
h, w = images.shape[1], images.shape[2]
img = np.zeros((h * size[0], w * size[1]))
for idx, image in enumerate(images):
i, j = int(idx % size[1]), int(idx / size[1])
image_ = imresize(image, size=(w, h), interp='bicubic')
img[j*h:(j+1)*h, i*w:(i+1)*w] = image_
return img
# scatter-plot
if args.scatter:
for i, (x_, y_) in enumerate([(xtr, ytr), (xte, yte)]):
model = TSNE(learning_rate=10)
mu, _ = sess.run(tennet, {x: x_})
results = model.fit_transform(mu)
plt.scatter(results[:,0], results[:,1],
c=['C%d' % np.argmax(y_[n,:]) for n in range(y_.shape[0])],
alpha=0.5)
handles = [mpatches.Patch(color='C%d'%c, label='%d'%c) for c in range(10)]
plt.legend(handles=handles)
flag = 'train' if i == 0 else 'test'
plt.savefig(os.path.join(visdir, 'scatter_%s.pdf'%flag), format='pdf')
plt.close()
# reconstruction
if args.reconstruct:
size = 5
idx = np.random.choice(range(1000), size=size**2, replace=False)
xtr_, xte_ = xtr[idx], xte[idx]
for (x_, flag) in [(xtr_,'[train]'), (xte_,'[test]')]:
# pass encoder and decoder
mu_z, _ = sess.run(tennet, {x: x_})
xmap = sess.run(tdenet, {z: mu_z})
xmap = xmap.reshape(size**2, 28, 28)
imsave(os.path.join(visdir, '%s reconstruct.png' % flag),
_merge(xmap, [size, size]))
x_ = x_.reshape(-1, 28, 28)
imsave(os.path.join(visdir, '%s original.png' % flag),
_merge(x_, [size, size]))
# 2-D manifold
if args.manifold:
zlim, size = 2, 20
z_ = np.mgrid[zlim:-zlim:size*1j, zlim:-zlim:size*1j]
z_ = np.rollaxis(z_, 0, 3).reshape([-1, 2])
xmap = sess.run(tdenet, {z: z_})
xmap = xmap.reshape(size*size, 28, 28)
imsave(os.path.join(visdir, 'manifold.png'), _merge(xmap, [size, size]))
if __name__=='__main__':
if args.mode == 'train':
train()
elif args.mode == 'test':
test()
elif args.mode == 'visualize':
visualize()
else:
raise ValueError('Invalid mode %s' % args.mode)