-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
134 lines (103 loc) · 5.82 KB
/
main.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
import os
import numpy as np
from sklearn.neighbors.kde import KernelDensity
import torch
from torch import nn, optim
from torch.utils.data import DataLoader
from torch.utils.data.sampler import SubsetRandomSampler
from torch.utils.tensorboard import SummaryWriter
from torchvision import datasets, transforms
from config import config
from models import GAN
from utils import load_checkpoints
def train(models, writer, device):
train_dataset = datasets.MNIST(config.dataset_dir, train=True, transform=transforms.ToTensor(), download=True)
val_dataset = datasets.MNIST(config.dataset_dir, train=True, transform=transforms.ToTensor())
indices = list(range(len(train_dataset)))
train_sampler = SubsetRandomSampler(indices[:55000])
val_sampler = SubsetRandomSampler(indices[55000:])
train_loader = DataLoader(train_dataset, batch_size=config.batch_size,
sampler=train_sampler, num_workers=config.num_workers)
val_loader = DataLoader(val_dataset, batch_size=config.batch_size,
sampler=val_sampler, num_workers=config.num_workers)
criterion = nn.BCEWithLogitsLoss()
gen_optimizer = optim.Adam(models.gen.parameters(), lr=config.gen_lr, betas=config.betas)
dis_optimizer = optim.Adam(models.dis.parameters(), lr=config.dis_lr, betas=config.betas)
idx = 0
real_label = torch.ones(1, device=device)
fake_label = torch.zeros(1, device=device)
for _ in range(config.train_epochs):
for data, _ in train_loader:
data = data.view(-1, 784).to(device)
# ----------------------------------------------------------------------------------------------------------
# Train the discriminator
# ----------------------------------------------------------------------------------------------------------
noise = torch.rand((data.size(0), config.noise_features), device=device) * 2 - 1
generated = models.gen(noise)
data_logits = models.dis(data)
generated_logits = models.dis(generated.detach())
real_labels = real_label.expand_as(data_logits)
fake_labels = fake_label.expand_as(data_logits)
dis_loss = criterion(data_logits, real_labels) + criterion(generated_logits, fake_labels)
dis_optimizer.zero_grad()
dis_loss.backward()
dis_optimizer.step()
# ----------------------------------------------------------------------------------------------------------
# Train the generator
# ----------------------------------------------------------------------------------------------------------
noise = torch.rand((data.size(0), config.noise_features), device=device) * 2 - 1
generated = models.gen(noise)
generated_logits = models.dis(generated)
gen_loss = criterion(generated_logits, real_labels)
gen_optimizer.zero_grad()
gen_loss.backward()
gen_optimizer.step()
if (idx + 1) % config.print_step == 0:
print("Iteration: {}/{}, gen_loss: {:.4f}, dis_loss: {:.4f}".format(
idx + 1, (55000 // config.batch_size + 1) * config.train_epochs, gen_loss.item(), dis_loss.item()))
if (idx + 1) % config.tensorboard_step == 0:
writer.add_scalar('Loss/generator', gen_loss.item(), idx + 1)
writer.add_scalar('Loss/discriminator', dis_loss.item(), idx + 1)
data = data.view(-1, 1, 28, 28)
generated = generated.view(-1, 1, 28, 28)
writer.add_images('Result/data', data, idx + 1)
writer.add_images('Result/generated', generated, idx + 1)
idx += 1
gen_path = os.path.join(config.checkpoint_dir, config.name, '{:02d}-gen.ckpt'.format(config.train_epochs))
dis_path = os.path.join(config.checkpoint_dir, config.name, '{:02d}-dis.ckpt'.format(config.train_epochs))
torch.save(models.gen.state_dict(), gen_path)
torch.save(models.dis.state_dict(), dis_path)
def test(models, device):
test_dataset = datasets.MNIST(config.dataset_dir, train=False, transform=transforms.ToTensor())
test_loader = DataLoader(test_dataset, batch_size=config.batch_size, num_workers=config.num_workers)
X_data = np.zeros((10000, 784), dtype=np.float32)
X_generated = np.zeros((10000, 784), dtype=np.float32)
with torch.no_grad():
for i, (data, _) in enumerate(test_loader):
noise = torch.rand((data.size(0), config.noise_features), device=device) * 2 - 1
generated = models.gen(noise)
start = i * config.batch_size
end = min((i + 1) * config.batch_size, 10000)
X_data[start:end] = data.view(-1, 784).numpy()
X_generated[start:end] = generated.cpu().numpy()
print("Calculating the score...")
kde = KernelDensity(bandwidth=0.2).fit(X_generated)
print("Score: {:.4f}".format(kde.score(X_data) / 10000))
def main():
if not os.path.exists(os.path.join(config.tensorboard_dir, config.name)):
os.makedirs(os.path.join(config.tensorboard_dir, config.name))
if not os.path.exists(os.path.join(config.checkpoint_dir, config.name)):
os.makedirs(os.path.join(config.checkpoint_dir, config.name))
device = torch.device('cuda:0' if config.use_cuda else 'cpu')
models = GAN(config).to(device)
if config.load_epoch != 0:
load_checkpoints(models, config.checkpoint_dir, config.name, config.load_epoch)
if config.is_train:
models.train()
writer = SummaryWriter(log_dir=os.path.join(config.tensorboard_dir, config.name))
train(models, writer, device)
else:
models.eval()
test(models, device)
if __name__ == '__main__':
main()