-
Notifications
You must be signed in to change notification settings - Fork 4
/
train.py
187 lines (147 loc) · 6.3 KB
/
train.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
from __future__ import print_function
from __future__ import division
import torch, time
import torch.utils.data
import matplotlib.pyplot as plt
from torchvision.utils import save_image, make_grid
from src.utils import *
from numpy.random import normal
def train_VAE(net, name, batch_size, nb_epochs, trainset, valset, cuda, flat_ims=False,
train_plot=False, Nclass=None, early_stop=None, script_mode=False):
models_dir = name + '_models'
results_dir = name + '_results'
mkdir(models_dir)
mkdir(results_dir)
if cuda:
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True, pin_memory=True,
num_workers=3)
valloader = torch.utils.data.DataLoader(valset, batch_size=batch_size, shuffle=False, pin_memory=True,
num_workers=3)
else:
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True, pin_memory=False,
num_workers=3)
valloader = torch.utils.data.DataLoader(valset, batch_size=batch_size, shuffle=False, pin_memory=False,
num_workers=3)
## ---------------------------------------------------------------------------------------------------------------------
# net dims
cprint('c', '\nNetwork:')
epoch = 0
## ---------------------------------------------------------------------------------------------------------------------
# train
cprint('c', '\nTrain:')
print(' init cost variables:')
vlb_train = np.zeros(nb_epochs)
vlb_dev = np.zeros(nb_epochs)
best_vlb = -np.inf
best_vlb_train = -np.inf
best_epoch = 0
nb_its_dev = 1
tic0 = time.time()
for i in range(epoch, nb_epochs):
net.set_mode_train(True)
tic = time.time()
nb_samples = 0
for x, y in trainloader:
if flat_ims:
x = x.view(x.shape[0], -1)
if Nclass is not None:
y_oh = torch_onehot(y, Nclass).type(x.type())
x = torch.cat([x, y_oh], 1)
cost, _ = net.fit(x)
vlb_train[i] += cost * len(x)
nb_samples += len(x)
vlb_train[i] /= nb_samples
toc = time.time()
# ---- print
print("it %d/%d, vlb %f, " % (i, nb_epochs, vlb_train[i]), end="")
cprint('r', ' time: %f seconds\n' % (toc - tic))
net.update_lr(i)
if vlb_train[i] > best_vlb_train:
best_vlb_train = vlb_train[i]
# ---- dev
if i % nb_its_dev == 0:
nb_samples = 0
for j, (x, y) in enumerate(valloader):
if flat_ims:
x = x.view(x.shape[0], -1)
if Nclass is not None:
y_oh = torch_onehot(y, Nclass).type(x.type())
x = torch.cat([x, y_oh], 1)
cost, _ = net.eval(x)
vlb_dev[i] += cost * len(x)
nb_samples += len(x)
vlb_dev[i] /= nb_samples
cprint('g', ' vlb %f (%f)\n' % (vlb_dev[i], best_vlb))
if train_plot:
zz = net.recongnition(x).sample()
o = net.regenerate(zz)
try:
o = o.cpu()
except:
o = o.loc.cpu()
if len(x.shape) == 2:
side = int(np.sqrt(x.shape[1]))
x = x.view(-1, 1, side, side).data
o = o.view(-1, 1, side, side).data
# save_image(torch.cat([x[:8], o[:8]]), results_dir + '/rec_%d.png' % i, nrow=8)
import matplotlib.pyplot as plt
plt.figure()
dd = make_grid(torch.cat([x[:10], o[:10]]), nrow=10).numpy()
plt.imshow(np.transpose(dd, (1, 2, 0)), interpolation='nearest')
if script_mode:
plt.savefig(results_dir + '/rec%d.png' % i)
else:
plt.show()
z_sample = normal(loc=0.0, scale=1.0, size=(36, net.latent_dim))
x_rec = net.regenerate(z_sample)
try:
x_rec = x_rec.cpu()
except:
x_rec = x_rec.loc.cpu()
if len(x_rec.shape) == 2:
side = int(np.sqrt(x_rec.shape[1]))
x_rec = x_rec.view(-1, 1, side, side)
plt.figure()
dd = make_grid(x_rec, nrow=6).numpy()
plt.imshow(np.transpose(dd, (1, 2, 0)), interpolation='nearest')
if script_mode:
plt.savefig(results_dir + '/sample%d.png' % i)
else:
plt.show()
if vlb_dev[i] > best_vlb:
best_vlb = vlb_dev[i]
best_epoch = i
net.save(models_dir + '/theta_best.dat')
if early_stop is not None and (i - best_epoch) > early_stop:
break
net.save(models_dir + '/theta_last.dat')
toc0 = time.time()
runtime_per_it = (toc0 - tic0) / float(nb_epochs)
cprint('r', ' average time: %f seconds\n' % runtime_per_it)
## ---------------------------------------------------------------------------------------------------------------------
# results
cprint('c', '\nRESULTS:')
nb_parameters = net.get_nb_parameters()
best_cost_dev = best_vlb
best_cost_train = best_vlb_train
print(' best_vlb_dev: %f' % best_cost_dev)
print(' best_vlb_train: %f' % best_cost_train)
print(' nb_parameters: %d (%s)\n' % (nb_parameters, humansize(nb_parameters)))
## ---------------------------------------------------------------------------------------------------------------------
# fig cost vs its
if not train_plot:
import matplotlib
matplotlib.use('agg')
import matplotlib.pyplot as plt
if train_plot:
plt.figure()
plt.plot(np.clip(vlb_train, -1000, 1000), 'r')
plt.plot(np.clip(vlb_dev[::nb_its_dev], -1000, 1000), 'b')
plt.legend(['cost_train', 'cost_dev'])
plt.ylabel('vlb')
plt.xlabel('it')
plt.grid(True)
plt.savefig(results_dir+'/train_cost.png')
if train_plot:
plt.show()
return vlb_train, vlb_dev