def main():
cwd = os.getcwd()
print(cwd)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
pretrained_net = models.resnet18(pretrained=True)
print(pretrained_net.fc)
pretrained_net.fc = nn.Linear(512, 2)
# 输出分类由imagenet的1000类，改为热狗数据集的2类
print(pretrained_net.fc)
# 分开设置lr，预训练部分模型的学习率0.01，fc部分的学习率0.1
output_params = list(map(id, pretrained_net.fc.parameters()))
feature_params = filter(lambda p: id(p) not in output_params,
pretrained_net.parameters())
lr = 0.001
optimizer = optim.SGD([{
'params': feature_params
}, {
'params': pretrained_net.fc.parameters(),
'lr': lr * 10
}],
lr=lr,
weight_decay=0.001)
batch_size = 128
num_epochs = 5
train_iter, test_iter = load_data_hotdog(batch_size)
trainer.train(pretrained_net, train_iter, test_iter, batch_size, optimizer,
device, num_epochs)
def main_no_pretrained():
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
scratch_net = models.resnet18(pretrained=False, num_classes=2)
lr = 0.1
optimizer = optim.SGD(scratch_net.parameters(), lr=lr, weight_decay=0.001)
batch_size = 128
num_epochs = 5
train_iter, test_iter = load_data_hotdog(batch_size)
trainer.train(scratch_net, train_iter, test_iter, batch_size, optimizer,
device, num_epochs)
def main():
cwd = os.getcwd()
print(cwd)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(net)
X = torch.rand((1, 1, 224, 224))
print(X, X.shape)
for name, blk in net.named_children():
X = blk(X)
print(name, 'output shape:', X.shape)
batch_size = 256
lr = 0.001
num_epochs = 5
# 在cifar10数据集上测试
train_iter, test_iter = cifar.load_data_cifar10(batch_size=batch_size)
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
trainer.train(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs)
FlattenLayer(),
nn.Linear(16 * 4 * 4, 120),
nn.BatchNorm1d(120),
nn.Sigmoid(),
nn.Linear(120, 84),
nn.BatchNorm1d(84),
nn.Sigmoid(),
nn.Linear(84, 10))
if __name__ == '__main__':
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
net = LeNet()
print(net)
batch_size = 256
train_iter, test_iter = fashion_mnist.load_data_fashion_mnist(
batch_size=batch_size)
lr = 0.001
num_epochs = 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
trainer.train(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs)
print("-" * 42)
print(lenet_bn)
optimizer = torch.optim.Adam(lenet_bn.parameters(), lr=lr)
train_iter, test_iter = fashion_mnist.load_data_fashion_mnist(
batch_size=batch_size)
trainer.train(lenet_bn, train_iter, test_iter, batch_size, optimizer,
device, num_epochs)
# 同ResNet一样，最后接上全局池化层和全连接层来输出。
net.add_module("BN", nn.BatchNorm2d(num_channels))
net.add_module("rele", nn.ReLU())
net.add_module(
"global_avg_pool",
GlobalAvgPool2d()) # GlobalAvgPool2d的输出（batch, num_channels, 1, 1）
net.add_module("fc", nn.Sequential(FlattenLayer(), nn.Linear(num_channels,
10))) # 类别数10
if __name__ == '__main__':
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(net)
X = torch.rand((1, 1, 96, 96))
for name, blk in net.named_children():
X = blk(X)
print(name, 'output shape:', X.shape)
batch_size = 256
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = fashion_mnist.load_data_fashion_mnist(
batch_size=batch_size, resize=96)
lr = 0.001
num_epochs = 5
model_path = "densenet_mnist.pt"
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
trainer.train(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs, model_path)
