def ResNet():
net = nn.Sequential(nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3),
nn.BatchNorm2d(64), nn.ReLU(),
nn.MaxPool2d(kernel_size=2, stride=2, padding=1))
#堆砌残差网络
def resnet_block(in_channels,
out_channels,
num_residuals,
first_block=False):
if first_block:
assert in_channels == out_channels
blk = []
for i in range(num_residuals):
if i == 0 and not first_block:
blk.append(
Residual(in_channels,
out_channels,
use_1x1conv=True,
stride=2))
else:
blk.append(Residual(out_channels, out_channels))
return nn.Sequential(*blk)
net.add_module('resnet_block1', resnet_block(64, 64, 2, first_block=True))
net.add_module('resnet_block2', resnet_block(64, 128, 2)) #通道增加 图像大小减半
net.add_module('resnet_block3', resnet_block(128, 256, 2))
net.add_module('resnet_block4', resnet_block(256, 512, 2))
net.add_module('gloval_ave_pool',
d2l.GlobalAvgPool2d()) #输出 (batch,512,1,1)
net.add_module('fc', nn.Sequential(d2l.FlattenLayer(), nn.Linear(512, 10)))
return net
def DenseNet():
#初始模块
net = nn.Sequential(nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3),
nn.BatchNorm2d(64), nn.ReLU(),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
num_channels, growth_rate = 64, 32
num_convs_in_dense_blocks = [4, 4, 4, 4]
# 稠密块 + 过渡块
for i, num_convs in enumerate(num_convs_in_dense_blocks):
DenseBlk = DenseBlock(num_convs, num_channels, growth_rate)
net.add_module('DenseBlock_%d' % i, DenseBlk)
# 稠密块的输出通道作为过渡层的输入
num_channels = DenseBlk.out_channels
#在稠密块(通道数增加) 之间加入过度层(图像大小减半，通道数减半)
if i != len(num_convs_in_dense_blocks) - 1:
TransBlk = transition_block(num_channels, num_channels // 2)
net.add_module('Trasition_block_%d' % i, TransBlk)
num_channels = num_channels // 2
net.add_module('BN', nn.BatchNorm2d(num_channels))
net.add_module('relu', nn.ReLU())
#GlobalAvgPool2d 输出 (Batch,num_channels,1,1)
net.add_module('global_avg_pool', d2l.GlobalAvgPool2d())
net.add_module(
'fc', nn.Sequential(d2l.FlattenLayer(), nn.Linear(num_channels, 10)))
return net
def get_net():
# 构建网络
# ResNet模型
model_path = r"F:\PyCharm\Practice\hand_wrtten\logs\Epoch100-Loss0.0000-train_acc1.0000-test_acc0.9930.pth"
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
net = nn.Sequential(nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3),
nn.BatchNorm2d(64), nn.ReLU(),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
net.add_module("resnet_block1", resnet_block(64, 64, 2, first_block=True))
net.add_module("resnet_block2", resnet_block(64, 128, 2))
net.add_module("resnet_block3", resnet_block(128, 256, 2))
net.add_module(
"global_avg_pool",
d2l.GlobalAvgPool2d()) # GlobalAvgPool2d的输出: (Batch, 512, 1, 1)
net.add_module("fc", nn.Sequential(d2l.FlattenLayer(), nn.Linear(256, 10)))
# 测试网络
# X = torch.rand((1, 1, 28, 28))
# for name, layer in net.named_children():
# X = layer(X)
# print(name, ' output shape:\t', X.shape)
# 加载网络模型
print("Load weight into state dict...")
stat_dict = torch.load(model_path, map_location=device)
net.load_state_dict(stat_dict)
net.to(device)
net.eval()
print("Load finish!")
return net
def vgg(conv_arch, fc_features, fc_hidden_units=4096):
net = nn.Sequential()
#Convolution layers
for i, (num_convs, in_channels, out_channels) in enumerate(conv_arch):
net.add_module("vgg_block_" + str(i + 1),
vgg_block(num_convs, in_channels, out_channels))
net.add_module(
"fc",
nn.Sequential(d2l.FlattenLayer(),
nn.Linear(fc_features, fc_hidden_units), nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(fc_hidden_units, fc_hidden_units), nn.ReLU(),
nn.Dropout(0.5), nn.Linear(fc_hidden_units, 10)))
return net
def vgg(conv_arch, fc_features, fc_hidden_units=4096):
net = nn.Sequential()
# 卷积层部分
for i, (num_convs, in_channels, out_channels) in enumerate(conv_arch): # enumerate() 函数用于将一个可遍历的数据对象(如列表、元组或字符串)组合为一个索引序列，同时列出数据和数据下标
# 每经过一个vgg_block都会使宽高减半
net.add_module("vgg_block_" + str(i+1), vgg_block(num_convs, in_channels, out_channels)) # 将一个child module 添加到当前modle, 被添加的module可以通过name属性来获取，
# 全连接层部分
net.add_module("fc", nn.Sequential(d2l.FlattenLayer(),
nn.Linear(fc_features, fc_hidden_units),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(fc_hidden_units, fc_hidden_units),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(fc_hidden_units, 10)
))
return net
def vgg(conv_arch, fc_features, fc_hidden_units=4096):
net = nn.Sequential()
# 卷积层部分
for i, (num_conv, in_channels, out_channels) in enumerate(conv_arch):
# 每经过一个vgg_block都会使宽高减半
net.add_module("vgg_block_" + str(i + 1),
vgg_block(num_conv, in_channels, out_channels))
# 全连接层
net.add_module(
"fc",
nn.Sequential(d2l.FlattenLayer(),
nn.Linear(fc_features, fc_hidden_units), nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(fc_hidden_units, fc_hidden_units), nn.ReLU(),
nn.Dropout(0.5), nn.Linear(fc_hidden_units, 10)))
return net
def vgg(conv_arch, fc_features, fc_hidden_units):
net = nn.Sequential()
# 卷积层部分
for i, (num_conv, in_channels, out_channels) in enumerate(conv_arch):
net.add_module(
'vgg_block_' + str(i + 1),
vgg_block(num_conv=num_conv,
in_channels=in_channels,
out_channels=out_channels))
# 全连接层部分
net.add_module(
'fc',
nn.Sequential(d2l.FlattenLayer(),
nn.Linear(fc_features, fc_hidden_units), nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(fc_hidden_units, fc_hidden_units), nn.ReLU(),
nn.Dropout(0.5), nn.Linear(fc_hidden_units, 10)))
return net
def vgg(conv_arch, fc_features, fc_hidden_unit=4096):
net = nn.Sequential()
# 卷积层部分
# enumerate()用于将一个可遍历的数据对象(如列表、元组或字符串)
# 组合为一个索引序列，同时列出数据和数据下标
for i, (num_convs, in_channels, out_channels) in enumerate(conv_arch):
net.add_module("vgg_blk_" + str(i + 1),
vgg_block(num_convs, in_channels, out_channels))
# 全连接层部分
net.add_module(
"fc",
nn.Sequential(
d2l.FlattenLayer(), # 四维转二维
nn.Linear(fc_features, fc_hidden_unit),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(fc_hidden_unit, fc_hidden_unit),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(fc_hidden_unit, 10)))
return net
def __simplified():
num_inputs, num_outputs, num_hiddens = 784, 10, 256
net = nn.Sequential(
d2l.FlattenLayer(),
nn.Linear(num_inputs, num_hiddens),
nn.ReLU(),
nn.Linear(num_hiddens, num_outputs),
)
for params in net.parameters():
init.normal_(params, mean=0, std=0.01)
batch_size = 256
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
loss = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(net.parameters(), lr=0.5)
# 训练模型
num_epochs = 5
d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size, None, None, optimizer)
def __init__(self):
super(VGG, self).__init__()
# conv_arch 有5个vgg block，每个block表示(卷积数，输入通道，输出通道)
self.conv_arch = ((1, 1, 64), (1, 64, 128), (2, 128, 256),
(2, 256, 512), (2, 512, 512))
self.fc_features = 512 * 7 * 7 # 经过5个conv 狂高减半5次 图片大小变为224／32 = 7
self.fc_hidden_unit = 4096
self.net = nn.Sequential()
#卷积
for i, (num_convs, in_channels,
out_channels) in enumerate(self.conv_arch):
self.net.add_module(
'vgg_block_' + str(i + 1),
self.vgg_block(num_convs, in_channels, out_channels))
#全链接
self.net.add_module(
'fc',
nn.Sequential(d2l.FlattenLayer(),
nn.Linear(self.fc_features, self.fc_hidden_unit),
nn.ReLU(), nn.Dropout(0.5),
nn.Linear(self.fc_hidden_unit, self.fc_hidden_unit),
nn.ReLU(), nn.Dropout(0.5),
nn.Linear(self.fc_hidden_unit, 10)))
b2 = nn.Sequential(nn.Conv2d(64, 64, kernel_size=1),
nn.Conv2d(64, 192, kernel_size=3, padding=1),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
b3 = nn.Sequential(Inception(192, 64, (96, 128), (16, 32), 32),
Inception(256, 128, (128, 192), (32, 96), 64),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
b4 = nn.Sequential(Inception(480, 192, (96, 208), (16, 48), 64),
Inception(512, 160, (112, 224), (24, 64), 64),
Inception(512, 128, (128, 256), (24, 64), 64),
Inception(512, 112, (144, 288), (32, 64), 64),
Inception(528, 256, (160, 320), (32, 128), 128),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
b5 = nn.Sequential(Inception(832, 256, (160, 320), (32, 128), 128),
Inception(832, 384, (192, 384), (48, 128), 128),
d2l.GlobalAvgPool2d())
net = nn.Sequential(b1, b2, b3, b4, b5, d2l.FlattenLayer(),
nn.Linear(1024, 10))
batch_size = 128
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=96)
lr, num_epochs = 0.001, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs)
feature = self.conv(X)
output = self.fc(feature)
return output
net = torch.nn.Sequential( #Lelet
Reshape(),
nn.Conv2d(in_channels=1, out_channels=6, kernel_size=5,
padding=2), #b*1*28*28 =>b*6*28*28
nn.Sigmoid(),
nn.AvgPool2d(kernel_size=2, stride=2), #b*6*28*28 =>b*6*14*14
nn.Conv2d(in_channels=6, out_channels=16,
kernel_size=5), #b*6*14*14 =>b*16*10*10
nn.Sigmoid(),
nn.AvgPool2d(kernel_size=2, stride=2), #b*16*10*10 => b*16*5*5
d2l.FlattenLayer(), #b*16*5*5 => b*400
nn.Linear(in_features=16 * 5 * 5, out_features=120),
nn.Sigmoid(),
nn.Linear(120, 84),
nn.Sigmoid(),
nn.Linear(84, 10))
LeNet = LeNet()
print(LeNet)
####################### AlexNet ##################
'''
LeNet: 在大的真实数据集上的表现并不尽如⼈意。
1.神经网络计算复杂。
2.还没有⼤量深⼊研究参数初始化和⾮凸优化算法等诸多领域。
for i, num_convs in enumerate(num_convs_in_dense_blocks):
DB = DenseBlock(num_convs, num_channels, growth_rate)
net.add_module("DenseBlosk_%d" % i, DB)
# 上一个稠密块的输出通道数
num_channels = DB.out_channels
# 在稠密块之间加入通道数减半的过渡层
if i != len(num_convs_in_dense_blocks) - 1:
net.add_module("transition_block_%d" % i, transition_block(num_channels, num_channels // 2))
num_channels = num_channels // 2
# 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("relu", nn.ReLU())
net.add_module("global_avg_pool", d2l.GlobalAvgPool2d()) # GlobalAvgPool2d的输出: (Batch, num_channels, 1, 1)
net.add_module("fc", nn.Sequential(d2l.FlattenLayer(), nn.Linear(num_channels, 10)))
'''
# 尝试打印每个子模块的输出维度确保网络无误：
X = torch.rand((1, 1, 96, 96))
for name, layer in net.named_children():
X = layer(X)
print(name, ' output shape:\t', X.shape)
'''
batch_size = 1
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=96)
lr, num_epochs = 0.001, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
def nin_block(in_channels, out_channels, kernel_size, stride, padding):
blk = nn.Sequential(
nn.Conv2d(in_channels, out_channels, kernel_size, stride, padding),
nn.ReLU(), nn.Conv2d(out_channels, out_channels, kernel_size=1),
nn.ReLU(), nn.Conv2d(out_channels, out_channels, kernel_size=1),
nn.ReLU())
return blk
net = nn.Sequential(nin_block(1, 96, kernel_size=11, stride=4, padding=0),
nn.MaxPool2d(kernel_size=3, stride=2),
nin_block(96, 256, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=3, stride=2),
nin_block(256, 384, kernel_size=3, stride=1, padding=1),
nn.MaxPool2d(kernel_size=3, stride=2), nn.Dropout(0.5),
nin_block(384, 10, kernel_size=3, stride=1, padding=1),
nn.AvgPool2d(kernel_size=5), d2l.FlattenLayer())
X = torch.rand(1, 1, 224, 224)
for name, blk in net.named_children():
X = blk(X)
print(name, 'output shape:', X.shape)
batch_size = 128
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)
lr, num_epochs = 0.002, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs)
self.moving_mean = self.moving_mean.to(device)
self.moving_var = self.moving_var.to(device)
Y, self.moving_mean, self.moving_var = batch_norm(self.training,
X,
self.gamma,
self.beta,
self.moving_mean,
self.moving_var,
eps=1e-5,
momentum=0.9)
return Y
net = nn.Sequential(nn.Conv2d(1, 6, 5), nn.Sigmoid(), nn.MaxPool2d(2, 2),
nn.Conv2d(6, 16, 5), BatchNorm(16, num_dims=4),
nn.Sigmoid(), nn.MaxPool2d(2, 2), d2l.FlattenLayer(),
nn.Linear(16 * 4 * 4, 120), BatchNorm(120, num_dims=2),
nn.Sigmoid(), nn.Linear(120, 84), BatchNorm(84,
num_dims=2),
nn.Sigmoid(), nn.Linear(84, 10))
batch_size = 64
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
lr, num_epochs = 0.001, 5
optimizer = optim.Adam(net.parameters(), lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs)
# print(net[1].gamma.view((-1,)), net[1].beta.view((-1,)))
def net(X, is_training=True):
X = X.view(-1, num_inputs)
H1 = (torch.matmul(X, W1) + b1).relu()
if is_training:
H1 = d2l.dropout(H1, drop_prob1) # 在第一层全连接后添加丢弃层（激活之后丢弃）
H2 = (torch.mm(H1, W2) + b2).relu()
if is_training:
H2 = d2l.dropout(H2, drop_prob2)
return torch.matmul(H2, W3) + b3
num_epochs, lr, batch_size = 4, 100.0, 256
loss = torch.nn.CrossEntropyLoss()
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
# 简洁版
net_pytorch = nn.Sequential(d2l.FlattenLayer(),
nn.Linear(num_inputs, num_hiddens1), nn.ReLU(),
nn.Dropout(drop_prob1),
nn.Linear(num_hiddens1, num_hiddens2), nn.ReLU(),
nn.Dropout(drop_prob2),
nn.Linear(num_hiddens2, num_outputs))
for p in params:
nn.init.normal_(p, mean=0, std=0.01)
# d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size, params, lr)
optimizer = torch.optim.SGD(net_pytorch.parameters(), lr=0.5)
d2l.train_ch3(net_pytorch, train_iter, test_iter, loss, num_epochs, batch_size,
None, None, optimizer)
else:
num_workers = 4
# load data quickly
train_iter = torch.utils.data.DataLoader(mnist_train, batch_size=batch_size, shuffle=True, num_workers=num_workers)
test_iter = torch.utils.data.DataLoader(mnist_test, batch_size=batch_size, shuffle=False, num_workers=num_workers)
return train_iter, test_iter
if __name__ == '__main__':
# dataset
num_inputs, num_outputs, num_hiddens = 784, 10, 256
batch_size = 256
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
# model
net = nn.Sequential(
d2l.FlattenLayer(), # change x shape
nn.Linear(num_inputs, num_hiddens),
nn.ReLU(),
nn.Linear(num_hiddens, num_outputs),
)
for params in net.parameters():
nn.init.normal_(params, mean=0, std=0.01)
loss = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(net.parameters(), lr=0.5)
# training
num_epochs = 5
d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size, None, None, optimizer)
n_train, n_test, true_w, true_b = 100, 100, [1.2, -3.4, 5.6], 5
features = torch.randn((n_train + n_test, 1))
poly_features = torch.cat((features, torch.pow(features, 2), torch.pow(features, 3)), 1)
labels = (true_w[0] * poly_features[:, 0] + true_w[1] * poly_features[:, 1]
drop_prob1, drop_prob2 = 0.2, 0.5
# is_training表示是否处于训练模式，评估模式不使用丢弃法
# def net(X, is_training = True):
# X = X.view(-1, num_inputs)
# H1 = (torch.matmul(X, W1) + b1).relu()
# if is_training:
# H1 = dropout(H1, drop_prob1)
# H2 = (torch.matmul(H1, W2) + b2).relu()
# if is_training:
# H2 = dropout(H2, drop_prob2)
# return torch.matmul(H2, W3) + b3
num_epochs, lr, batch_size = 5, 100.0, 256
loss = torch.nn.CrossEntropyLoss()
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
# d2l.train_ch3(net,train_iter,test_iter,loss,num_epochs,batch_size,params,lr)
# 简洁实现
net = nn.Sequential(d2l.FlattenLayer(), nn.Linear(num_inputs, num_hiddens1),
nn.ReLU(), nn.Dropout(drop_prob1),
nn.Linear(num_hiddens1, num_hiddens2), nn.ReLU(),
nn.Dropout(drop_prob2), nn.Linear(num_hiddens2,
num_outputs))
for param in net.parameters():
nn.init.normal_(param, mean=0, std=0.01)
optimizer = torch.optim.SGD(net.parameters(), lr=0.5)
d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size, None,
None, optimizer)
self.moving_var = self.moving_var.to(X.device)
Y, self.moving_mean, self.moving_var = batch_norm(self.training,
X,
self.gamma,
self.beta,
self.moving_mean,
self.moving_var,
eps=1e-5,
momentum=0.9)
return Y
net = nn.Sequential(nn.Conv2d(1, 6, 5), BatchNorm(6, num_dims=4), nn.Sigmod(),
nn.MaxPool2d(2, 2), nn.Conv2d(6, 16, 5),
BatchNorm(16, num_dims=4), nn.Sigmod(), nn.MaxPool2d(2, 2),
d2l.FlattenLayer(), nn.Linear(16 * 4 * 4, 120),
BatchNorm(120,
num_dims=2), nn.Sigmoid(), nn.Linear(120, 84),
BatchNorm(84, num_dims=2), nn.Sigmoid(), nn.Linear(84, 10))
batch_size = 256
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size=batch_size)
lr, num_epochs = 0.001, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, otpimizer, device,
num_epochs)
net[1].gamma.view((-1)), net[1].beta.view((-1, ))
net = nn.Sequential(nn.Conv2d(1, 6, 5), nn.BatchNorm2d(6), nn.Sigmoid(),
nn.MaxPool2d(2, 2), nn.Conv2d(6, 16, 5),
def __init__(self, num_inputs, num_hiddens, num_outputs):
super(mlp, self).__init__()
self.FlattenLayer = d2l.FlattenLayer()
self.Linear1 = nn.Linear(num_inputs, num_hiddens)
self.ReLU = nn.ReLU()
self.Linear2 = nn.Linear(num_hiddens, num_outputs)
blk = []
for i in range(num_residuals):
if i == 0 and not first_block:
blk.append(
Residual(in_channels, out_channels, use_1x1conv=True,
stride=2))
else:
blk.append(Residual(out_channels, out_channels))
return nn.Sequential(*blk)
# 对第一个模块进行特殊处理，输入通道数等于输出通道数，而后续的模块则将通道数翻倍
net.add_module("resnet_block1", resnet_block(64, 64, 2, first_block=True))
net.add_module("resnet_block2", resnet_block(64, 128, 2))
net.add_module("resnet_block3", resnet_block(128, 256, 2))
net.add_module("resnet_block4", resnet_block(256, 512, 2))
# 全局平均池化后接上全连接层
net.add_module("global_avg_pool",
d2l.GlobalAvgPool2d()) # GlobalAvgPool2d的输出: (Batch, 512, 1, 1)
net.add_module("fc", nn.Sequential(d2l.FlattenLayer(), nn.Linear(512, 10)))
batch_size = 2
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=96)
lr, num_epochs = 0.001, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs)
# 定义NiN模型
# 不使用全连接层，通过卷积层是输出通道数等于类别标签数
# 减小模型参数尺寸，缓解过拟合，但有时会造成训练时间增加
net = nn.Sequential(
nin_block(1, 96, kernel_size=11, stride=4, padding=0), # (224-11+4-1)/4=54
nn.MaxPool2d(kernel_size=3, stride=2), # (54-1)/2=26
nin_block(96, 256, kernel_size=5, stride=1, padding=2), # 不变
nn.MaxPool2d(kernel_size=3, stride=2), # (26-1)/2=12
nin_block(256, 384, kernel_size=3, stride=1, padding=1), # 不变
nn.MaxPool2d(kernel_size=3, stride=2), # (12-1)/2=5
nn.Dropout(0.5), # 丢弃法
nin_block(384, 10, kernel_size=3, stride=1, padding=1), # 类别标签数是10
d2l.GlobalAvgPool2d(), # 全局平均池化层 (batch_size, 10, 1, 1)
d2l.FlattenLayer() # 维度转化 (batch_size, 10)
)
# 测试
# x = torch.rand(1,1,224,224)
# for name, blk in net.named_children(): # name_children()返回每一模块的名字和对象
# x = blk(x)
# print(name, 'output shape: ', x.shape)
# 参数
lr, num_epochs, batch_size = 0.002, 5, 64
# 数据
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)
# 优化器
optimizer = optim.Adam(net.parameters(), lr=lr)
# 训练
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device,
print (dropout(X, 0.5))
print (dropout(X, 1.0))
'''
num_inputs, num_outputs, num_hiddens1, num_hiddens2 = 784, 10, 256, 256
W1 = torch.tensor(np.random.normal(0, 0.01, size = (num_inputs, num_hiddens1)), dtype = torch.float, requires_grad = True)
b1 = torch.zeros(num_hiddens1, requires_grad = True)
W2 = torch.tensor(np.random.normal(0, 0.01, size = (num_hiddens1, num_hiddens2)), dtype = torch.float, requires_grad = True)
b2 = torch.zeros(num_hiddens2, requires_grad = True)
W3 = torch.tensor(np.random.normal(0, 0.01, size = (num_hiddens2, num_outputs)), dtype = torch.float, requires_grad = True)
b3 = torch.zeros(num_outputs, requires_grad = True)
params = [W1, b1, W2, b2, W3, b3]
drop_prob1, drop_prob2 = 0.2, 0.5
net = nn.Sequential(
d2l.FlattenLayer(),#对X的形状转换
nn.Linear(num_inputs, num_hiddens1),
nn.ReLU(),
nn.Dropout(drop_prob1),
nn.Linear(num_hiddens1, num_hiddens2),
nn.ReLU(),
nn.Dropout(drop_prob2),
nn.Linear(num_hiddens2, num_outputs)
)
for param in net.parameters():
nn.init.normal_(param, mean = 0, std = 0.01)
optimizer = torch.optim.SGD(net.parameters(), lr = 0.5)
num_epochs, lr, batch_size = 5, 100.0, 256
loss = torch.nn.CrossEntropyLoss()
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, batch_size, None, None, optimizer)
class GlobalAvgPool2d(nn.Module):
def __init__(self):
super(GlobalAvgPool2d, self).__init__()
def forward(self, x):
return F.avg_pool2d(x, kernel_size=x.size()[2:])
net = nn.Sequential(nin_block(1, 96, kernel_size=11, stride=4, padding=0),
nn.MaxPool2d(kernel_size=3, stride=2),
nin_block(96, 256, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=3, stride=2),
nin_block(256, 384, kernel_size=3, stride=1, padding=1),
nn.MaxPool2d(kernel_size=3, stride=2), nn.Dropout(0.5),
nin_block(384, 10, kernel_size=3, stride=1, padding=1),
GlobalAvgPool2d(), d2l.FlattenLayer())
X = t.rand(1, 1, 224, 224)
for name, blk in net.named_children():
X = blk(X)
print(name, "output shape : ", X.shape)
batch_size = 128
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)
lr, num_epochs = 0.002, 5
optimizer = t.optim.Adam(net.parameters(), lr=lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs)
Inception(528, 256, (160, 320), (32, 128), 128),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
# In[7]:
b5 = nn.Sequential(Inception(832, 256, (160, 320), (32, 128), 128),
Inception(832, 384, (192, 384), (48, 128), 128),
d2l.GlobalAvgPool2d())
# In[8]:
net = nn.Sequential(b1, b2, b3, b4, b5, d2l.FlattenLayer(), nn.Linear(1024, 10))
X = torch.rand(1, 1, 96, 96)
for blk in net.children():
X = blk(X)
print('output shape: ', X.shape)
# ## 5.9.3 获取数据和训练模型
# In[9]:
batch_size = 128
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=96)
net.add_module("DenseBlock_%d" % i, DB)
# 上一个稠密块的输出通道数
num_channels = DB.out_channels
# 在稠密块之间加入通道减半的过渡层
if i != len(num_convs_in_dense_blocks) - 1:
net.add_module("transition_block_%d" % i,
transition_block(num_channels, num_channels // 2))
num_channels = num_channels // 2
# 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("relu", nn.ReLU())
# GlobalAvgPool2d的输出: (Batch, num_channels, 1, 1)
net.add_module("global_avg_pool", d2l.GlobalAvgPool2d())
net.add_module("fc",
nn.Sequential(d2l.FlattenLayer(), nn.Linear(num_channels, 10)))
# 我们尝试打印每个子模块的输出维度确保网络无误：
X = torch.rand((1, 1, 96, 96))
for name, layer in net.named_children():
X = layer(X)
print(name, ' output shape:\t', X.shape)
blk = DenseBlock(2, 3, 10)
X = torch.rand(4, 3, 8, 8)
Y = blk(X)
print(Y.shape)
# summary(blk, (3, 8, 8))
blk = transition_block(23, 10)
print(blk(Y).shape)
device = torch.device(
torch.device('cuda:0' if torch.cuda.is_available() else 'cpu'))
# device = torch.device(torch.device('cpu'))
## Load data.
batch_size = 256
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
num_inputs = 28 * 28
num_outputs = 10
## Define model.
net = nn.Sequential(
# FlattenLayer()
# nn.Linear(num_inputs, num_outputs)
OrderedDict([('flatten', d2l.FlattenLayer()),
('linear', nn.Linear(num_inputs, num_outputs))])).to(device)
## Init paras.
init.normal_(net.linear.weight, mean=0, std=0.01)
init.constant_(net.linear.bias, val=0)
## Define loss function.
loss = nn.CrossEntropyLoss().to(device)
## Define optimization function.
optimizer = torch.optim.SGD(net.parameters(), lr=0.1)
## Train model.
num_epoches = 5
d2l.train_ch3(net, train_iter, test_iter, loss, num_epoches, batch_size, None,
3.10 多层感知机的简洁实现
https://tangshusen.me/Dive-into-DL-PyTorch/#/chapter03_DL-basics/3.10_mlp-pytorch
'''
import torch
from torch import nn
from torch.nn import init
import numpy as np
import sys
sys.path.append("..")
import d2lzh_pytorch as d2l
num_inputs, num_outputs, num_hiddens = 784, 10, 256
net = nn.Sequential(
d2l.FlattenLayer(),
nn.Linear(num_inputs, num_hiddens),
nn.ReLU(),
nn.Linear(num_hiddens, num_outputs),
)
for params in net.parameters():
init.normal_(params, mean=0, std=0.01)
batch_size = 256
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
loss = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(net.parameters(), lr=0.5)
num_epochs = 5
"""
import torch
import torchvision
import numpy as np
import sys
from torch import nn
from collections import OrderedDict
from torch.nn import init
import d2lzh_pytorch as d2l
# 定义模型
# 神经网络,用来求导，计算梯度
num_inputs, num_outputs = 784, 10
net = nn.Sequential()
net.add_module('flatten', d2l.FlattenLayer())
net.add_module('linear', nn.Linear(num_inputs, num_outputs))
# 初始化参数(但其实初始化网络时，参数已经初始化)
init.normal_(net.linear.weight, mean=0, std=0.01)
init.constant_(net.linear.bias, val=0)
# 定义损失函数
# PyTorch提供了一个包括softmax运算的交叉熵损失函数
loss = nn.CrossEntropyLoss()
# 定义最优化算法,用来更新参数
optimizer = torch.optim.SGD(net.parameters(), lr=0.1)
# 初始化参数
batch_size, num_epochs = 256, 5