-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain.py
214 lines (184 loc) · 8.63 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
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
"""Training procedure for NICE.
"""
import argparse
import torch, torchvision
from torchvision import transforms
import numpy as np
import nice
import time
import pickle
import matplotlib.pyplot as plt
def train(flow, trainloader, optimizer, epoch, device, sample_shape, filename, train_loss):
running_loss = 0
batches = 0
start = time.time()
flow.train() # set to training mode
for batch_idx, (inputs, _) in enumerate(trainloader):
inputs = inputs.view(inputs.shape[0], inputs.shape[1] * inputs.shape[2] * inputs.shape[
3]) # change shape from BxCxHxW to Bx(C*H*W)
# TODO Fill in
optimizer.zero_grad()
inputs = inputs.to(device)
loss = -flow(inputs).mean() # over batch, minus for minimize instead maximize the log_porb
loss.backward()
optimizer.step()
#print(f"batch: {loss}")
# print statistics
running_loss += loss.item()
batches += 1
end = time.time()
print(f" Train: epoch: {epoch},\t | time: {end - start} \n "
f"loss: {running_loss/batches}")
train_loss.append(running_loss/batches)
flow.eval() # set to inference mode
with torch.no_grad():
#reconstruction
z, _ = flow.f(inputs)
recont = flow.f_inverse(z).cpu()
recont = recont.view(-1, sample_shape[0], sample_shape[1], sample_shape[2]) # convet to BxCxHxW
torchvision.utils.save_image(torchvision.utils.make_grid(recont),
'./reconstruction/' + filename + 'epoch%d.png' % epoch)
def test(flow, testloader, filename, epoch, sample_shape, device, test_loss):
flow.eval() # set to inference mode
with torch.no_grad():
samples = flow.sample(100).cpu()
samples = samples.view(-1, sample_shape[0], sample_shape[1], sample_shape[2]) # convet to BxCxHxW
torchvision.utils.save_image(torchvision.utils.make_grid(samples),
'./samples/' + filename + 'epoch%d.png' % epoch)
# TODO full in
running_loss = 0
batches = 0
for batch_idx, (inputs, _) in enumerate(testloader):
batches += 1
inputs = inputs.view(inputs.shape[0], inputs.shape[1] * inputs.shape[2] * inputs.shape[
3]) # change shape from BxCxHxW to Bx(C*H*W)
inputs = inputs.to(device)
loss = -flow(inputs).mean() # over batch, minus for minimize instead maximize the log_porb
# print statistics
running_loss += loss.item()
print(f" Test: epoch - {epoch} \n loss: {running_loss / batches}")
test_loss.append(running_loss / batches)
def main(args):
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
sample_shape = [1, 28, 28]
transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.5,), (1.,)),
transforms.Lambda(lambda x: x + torch.zeros_like(x).uniform_(0., 1. / 256.)) # dequantization
])
if args.dataset == 'mnist':
trainset = torchvision.datasets.MNIST(root='./data/MNIST',
train=True, download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset,
batch_size=args.batch_size, shuffle=True, num_workers=2)
testset = torchvision.datasets.MNIST(root='./data/MNIST',
train=False, download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset,
batch_size=args.batch_size, shuffle=False, num_workers=2)
full_dim = 1 * 28 * 28 # CxHxW
elif args.dataset == 'fashion-mnist':
trainset = torchvision.datasets.FashionMNIST(root='~/torch/data/FashionMNIST',
train=True, download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset,
batch_size=args.batch_size, shuffle=True, num_workers=2)
testset = torchvision.datasets.FashionMNIST(root='./data/FashionMNIST',
train=False, download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset,
batch_size=args.batch_size, shuffle=False, num_workers=2)
full_dim = 1 * 28 * 28 # CxHxW
else:
raise ValueError('Dataset not implemented')
model_save_filename = '%s_' % args.dataset \
+ 'batch%d_' % args.batch_size \
+ 'coupling%s_' % args.coupling_name \
+ 'mid%d_' % args.mid_dim \
+ 'hidden%d_' % args.hidden
flow = nice.NICE(
prior=args.prior,
coupling_name=args.coupling_name,
num_coupling=args.num_coupling,
in_out_dim=full_dim,
mid_dim=args.mid_dim,
hidden=args.hidden,
device=device).to(device)
optimizer = torch.optim.Adam(
flow.parameters(), lr=args.lr) #, betas=[args.beta1, args.beta2], eps=args.ep, weight_decay=1)
#my_lr_scheduler = torch.optim.lr_scheduler.ExponentialLR(optimizer=optimizer, gamma=0.94)
train_loss = []
test_loss = []
for epoch_idx in range(args.epochs):
train(flow=flow, trainloader=trainloader,
optimizer=optimizer, epoch=epoch_idx,
device=device, sample_shape=sample_shape,
filename=model_save_filename, train_loss=train_loss)
#my_lr_scheduler.step()
test(flow=flow, testloader=testloader, filename=model_save_filename, epoch=epoch_idx,
sample_shape=sample_shape, device=device, test_loss=test_loss)
train_loss = np.array(train_loss)
test_loss = np.array(test_loss)
np.savez(f"train_loss_{args.dataset}_{args.coupling_name}", train=train_loss)
np.savez(f"test_loss_{args.dataset}_{args.coupling_name}", test=test_loss)
torch.save({
'num_epoch': args.epochs,
'model_state_dict': flow.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
'dataset': args.dataset,
'batch_size': args.batch_size,
'prior': args.prior,
'coupling_name': args.coupling_name,
'mid_dim': args.mid_dim,
'hidden': args.hidden, },
'./models/' + model_save_filename + '.tar')
print('Checkpoint Saved')
#save plot
plt.plot(np.arange(1, 51), test_loss, np.arange(1, 51), train_loss)
plt.xlabel('Epoch')
plt.ylabel('Loss = - Log-likelihood')
plt.title(f'Loss_Vs_Epoch - {args.dataset}_{args.coupling_name}')
plt.grid(True)
plt.legend(['Test', 'Train'], loc='upper right')
plt.savefig(f'loss_vs_epoch_{args.dataset}_{args.coupling_name}.jpg')
if __name__ == '__main__':
parser = argparse.ArgumentParser('')
parser.add_argument('--dataset',
help='dataset to be modeled.',
type=str,
default='mnist')
parser.add_argument('--prior',
help='latent distribution.',
type=str,
default='logistic')
parser.add_argument('--batch_size',
help='number of images in a mini-batch.',
type=int,
default=128)
parser.add_argument('--epochs',
help='maximum number of iterations.',
type=int,
default=50)
parser.add_argument('--sample_size',
help='number of images to generate.',
type=int,
default=64)
parser.add_argument('--num_coupling',
help='num of coupling layers',
type=int,
default='4')
parser.add_argument('--coupling_name',
help='.',
type=str,
default='additive')
parser.add_argument('--mid-dim',
help='.',
type=int,
default=1000)
parser.add_argument('--hidden',
help='.',
type=int,
default=5)
parser.add_argument('--lr',
help='initial learning rate.',
type=float,
default=1e-3)
args = parser.parse_args()
main(args)