LeNet
- LeNet的结构如下,它的输入是一个32 * 32的image,放到一个5 * 5的卷积层中,它的通道数是6,它的输出是一个28 * 28的一个矩阵,这里有个概念叫做feature map,也就第卷积层的输出的那一坨就是feature map。接下来使用一个2*2的pooling层,这样就从28 * 28转换为了14 * 14,通道数目不变。接下来又是一个5 * 5的卷积层,这输出从14 * 14转换为了10 * 10,通道数变为16。后面再接上一个pooling层,输出为一个5 * 5,通道数为16。然后将最后一个pooling层接到一个全连接层,全连接层的输入是最后一个pooling层得到的一坨拉成一个向量的长度为5 * 5 * 16, 他的输出的向量长度为120,再输入一个全连接层,这个全连接层的输出是64,最后接一个Gauss层(现在不用了,基本上可以认为是一个softmax层)。
总结
LeNet的实现
LeNet(LeNet-5)由两个部分组成:卷积编码器和全连接层密集块
import torch
import torchvision
from d2l import torch as d2l
from torch import nn
from torch.utils import data
from torchvision import transforms
# 这个类的作用就是将我们的额输入X变为一个批量数不变,通道数变为1,大小变为28*28
class Reshape(nn.Module):
def forward(self, x):
return x.view(-1, 1, 28, 28)
LeNet = nn.Sequential(
Reshape(), nn.Conv2d(1, 6, kernel_size=5, padding=2), nn.Sigmoid(), # 第一个卷积层,经过Reshape()后我们得到的输入数据是28*28的大小,这里我们对数据做一个填充使它变为32*32大小,这里我们的卷积核的大小为5*5
nn.AvgPool2d(kernel_size=2, stride=2), # 第一个池化层,经过上述的卷积层,我这一层的输入的形状是6*28*28,这里我们选择的池化窗口的大小是2*2,步幅是2,步幅和窗口大小相同,也就是不重复扫描,这一层的输出的形状为6*14*4
nn.Conv2d(6, 16, kernel_size=5), nn.Sigmoid(), # 第二个卷积层,经过上面的池化层,这里的输入的形状是6*14*14, 输出通道的大小为16,卷积核的大小是5*5,所以输出形状是16*10*10
nn.AvgPool2d(kernel_size=2, stride=2), nn.Flatten(), # 第二个池化层,上面的卷积层后,这里的输入是16*10*10,这里选择的池化窗口的大小是6,步幅为2,所以我们得到输出的形状是16*5*5,然后使用Flatten(),将它展平,也就是保持维度0不变,其他的维度拉直
nn.Linear(16 * 5 * 5, 120), nn.Sigmoid(), # 第一个全连接层,上面的操作我们得到了一个长度为16*5*5的向量,作为该层输入,这一层的输出是长度为120的向量,激活函数选用Sigmoid
nn.Linear(120, 84), nn.Sigmoid(), # 第二个全连接层,这一层的输入是上一层的输出,输出是一个长度为84的向量,激活函数选用Sigmoid
nn.Linear(84, 10)) # 第三个全连接层,输入是上一层的输出,输出是长度为10的向量
# 检查模型
X = torch.rand(size=(1, 1, 28, 28), dtype=torch.float32)
for layer in LeNet:
X = layer(X)
print(layer.__class__.__name__, "output shape:\t", X.shape)
Reshape output shape: torch.Size([1, 1, 28, 28])
Conv2d output shape: torch.Size([1, 6, 28, 28])
ReLU output shape: torch.Size([1, 6, 28, 28])
AvgPool2d output shape: torch.Size([1, 6, 14, 14])
Conv2d output shape: torch.Size([1, 16, 10, 10])
ReLU output shape: torch.Size([1, 16, 10, 10])
AvgPool2d output shape: torch.Size([1, 16, 5, 5])
Flatten output shape: torch.Size([1, 400])
Linear output shape: torch.Size([1, 120])
ReLU output shape: torch.Size([1, 120])
Linear output shape: torch.Size([1, 84])
ReLU output shape: torch.Size([1, 84])
Linear output shape: torch.Size([1, 10])
# 测试LeNet在Fashion-MNIST数据集上的表现
batch_size = 256
trans = transforms.ToTensor()
train_mnist = torchvision.datasets.FashionMNIST(root="./fashion-mnist", train=True, transform=trans, download=True)
test_mnist = torchvision.datasets.FashionMNIST(root="./fashion-mnist", train=False, transform=trans, download=True)
train_iter = data.DataLoader(train_mnist, batch_size=batch_size)
test_iter = data.DataLoader(test_mnist, batch_size=batch_size)
/Users/tiger/opt/anaconda3/envs/d2l-zh/lib/python3.8/site-packages/torchvision/datasets/mnist.py:498: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at ../torch/csrc/utils/tensor_numpy.cpp:180.)
return torch.from_numpy(parsed.astype(m[2], copy=False)).view(*s)
# evaluate_accuracy函数
def evaluate_accuracy(net, data_iter, device=None):
"""使用GPU计算模型在数据集上的精度"""
if isinstance(net, torch.nn.Module):
net.eval() # eval的意识就是关闭模型中的dropout功能,调到评价模式;与之相对应的是train()
if not device:
device = next(iter(net.parameters())).device
metric = d2l.Accumulator(2)
for X, y in data_iter:
if isinstance(X, list):
X = [x.to(device) for x in X]
else:
X = X.to(device)
y = y.to(device)
y = y.to(device)
metric.add(d2l.accuracy(net(X), y), y.numel())
return metric[0] / metric[1 ]
def train_ch6(net, train_iter, test_iter, num_epochs, lr, device):
"""用GPU训练模型(在第六章定义)。"""
# 初始化weight
def init_weights(m):
if type(m) == nn.Linear or type(m) == nn.Conv2d:
nn.init.xavier_uniform_(m.weight)
net.apply(init_weights)
print('training on', device)
# 将网络挪到对应的设备
net.to(device)
optimizer = torch.optim.SGD(net.parameters(), lr=lr)
loss = nn.CrossEntropyLoss()
animator = d2l.Animator(xlabel='epoch', xlim=[1, num_epochs],legend=['train loss', 'train acc', 'test acc'])
timer, num_batches = d2l.Timer(), len(train_iter)
for epoch in range(num_epochs):
metric = d2l.Accumulator(3)
net.train()
for i, (X, y) in enumerate(train_iter):
timer.start()
optimizer.zero_grad()
# 每一批数据挪到对应的设备
X, y = X.to(device), y.to(device)
y_hat = net(X)
l = loss(y_hat, y)
l.backward()
optimizer.step()
with torch.no_grad():
metric.add(l * X.shape[0], d2l.accuracy(y_hat, y), X.shape[0])
timer.stop()
train_l = metric[0] / metric[2]
train_acc = metric[1] / metric[2]
if (i + 1) % (num_batches // 5) == 0 or i == num_batches - 1:
animator.add(epoch + (i + 1) / num_batches,
(train_l, train_acc, None))
test_acc = evaluate_accuracy(net, test_iter)
animator.add(epoch + 1, (None, None, test_acc))
print(f'loss {train_l:.3f}, train acc {train_acc:.3f}, 'f'test acc {test_acc:.3f}')
print(f'{metric[2] * num_epochs / timer.sum():.1f} examples/sec 'f'on {str(device)}')
# 训练和评估LeNet-5模型
lr, num_epochs = 0.1, 10
train_ch6(LeNet, train_iter, test_iter, num_epochs, lr, "cpu")
loss 0.348, train acc 0.872, test acc 0.859
6905.4 examples/sec on cpu