实战配套视频：《PyTorch深度学习实践》完结合集
实战笔记：
Learning_AI的博客_学习CV的研一小白_PyTorch学习笔记,
刘二大人：pytorch深度学习实践（代码详细笔记，适合零基础）
pytorch实战教学（一篇管够）_小星AI-CSDN博客_pytorch实战
Pytorch学习笔记--Bilibili刘二大人Pytorch教学代码汇总
目录
【深度学习|Pytorch深度学习实战笔记】线性模型
梯度下降
随机梯度下降：
反向传播
Pytorch实战--线性回归
Pytorch实战--逻辑回归
处理多维特征的输入
加载数据集
多分类问题
卷积神经网络
卷积神经网络（高级）
Residual net残差结构块
RNN
线性模型
import numpy as np
import matplotlib.pyplot as pltx_data = https://www.it610.com/article/[1.0, 2.0, 3.0]
y_data = [2.0, 4.0, 6.0]def forward(x):
return x * wdef loss(x, y):
y_pred = forward(x)
return (y_pred - y) * (y_pred - y)w_list = []
mse_list = []
for w in np.arange(0.0, 4.1, 0.1):
print('w=', w)
l_sum = 0
for x_val, y_val in zip(x_data, y_data):
y_pred_val = forward(x_val)
loss_val = loss(x_val, y_val)
l_sum += loss_val
print('\t', x_val, y_val, y_pred_val, loss_val)
print('MSE=', l_sum / 3)
w_list.append(w)
mse_list.append(l_sum / 3)plt.plot(w_list, mse_list)
plt.ylabel('Loss')
plt.xlabel('w')
plt.show()
运行截图如下：
文章图片
梯度下降 以模型
文章图片
为例，梯度下降算法就是一种训练参数
文章图片
到最佳值的一种算法，
文章图片
每次变化的趋势由
文章图片
（学习率：一种超参数，由人手动设置调节），以及
文章图片
的导数来决定，具体公式如下：
文章图片
注： 此时
文章图片
函数是指所有的损失函数之和
针对模型
文章图片
的梯度下降算法的公式化简如下：
文章图片
# 输入训练数据
x_data = https://www.it610.com/article/[1.0, 2.0, 3.0]
y_data = [2.0, 4.0, 6.0]
# 设置初始参数
w = 1.0# 初始权重
alpha = 0.005 #初始梯度下降法的学习率
# 定义计算y_hat的函数
def forward(x):
return x * w
# 定义计算平均损失的函数
def cost(xs, ys):
sum_cost = 0
for x, y in zip(xs, ys):# zip函数的功能是打包为元组列表
y_pred = forward(x)
sum_cost += (y_pred - y) ** 2
return sum_cost / len(xs)
def gradient(xs, ys):
grad = 0
for x, y in zip(xs, ys):
grad += 2 * x * (x * w - y)
return grad / len(xs)
print('Predict (before training)', 4, forward(4))# 计算训练前初始参数对应的y_hat值
for epoch in range(1000):
cost_val = cost(x_data, y_data)# 计算平均损失值
grad_val = gradient(x_data, y_data) # 计算梯度值
w -= alpha * grad_val # 更新权重w
print('Epoch', epoch, 'w = ', w, 'loss = ', cost_val) # 输出当前迭代次数的权重值和平均损失值
print('Predict (after training)', 4, forward(4)) #计算训练权重w后，对应的y_hat值
随机梯度下降： 随机梯度下降算法与梯度下降算法的不同之处在于，随机梯度下降算法不再计算损失函数之和的导数，而是随机选取任一随机函数计算导数，随机的决定
文章图片
下次的变化趋势，具体公式变化如图：
文章图片
# 输入训练数据
x_data = https://www.it610.com/article/[1.0, 2.0, 3.0]
y_data = [2.0, 4.0, 6.0]
# 设置初始参数
w = 1.0# 初始权重
alpha = 0.005 #初始梯度下降法的学习率
# 定义计算y_hat的函数
def forward(x):
return x * w
# 定义计算单个样本损失的函数
def loss(xs, ys):
y_pred = forward(x) # 计算预测值y_hat
single_lost = (y_pred - ys) ** 2 # 计算误差
return single_lost
def gradient(xs, ys):
grad = 2 * x * (x * w - y)
return grad
print('Predict (before training)', 4, forward(4))# 计算训练前初始参数对应的y_hat值
for epoch in range(1000): # 迭代次数
for x, y in zip(x_data, y_data): # 遍历数据
grad_val = gradient(x, y)# 计算当前数据的梯度值
w -= alpha * grad_val# 更新权重w
print("\tgrad: ", x, y, grad_val)
los = loss(x, y)# 计算当前数据的损失值
print('progress: ', epoch, 'w = ', w, 'loss = ', los)
print('Predict (after training)', 4, forward(4)) # 计算训练权重w后，对应的y_hat值反向传播
import torchx_data = https://www.it610.com/article/[1.0, 2.0, 3.0]
y_data = [2.0, 4.0, 6.0]w = torch.Tensor([1.0])
w.requires_grad = True# 需要计算梯度def forward(x):
return x * w# tensordef loss(x, y):
y_pred = forward(x)
return (y_pred - y) ** 2print('predict (before training)', 4, forward(4).item())
for epoch in range(100):
for x, y in zip(x_data, y_data):
l = loss(x, y) # 前向，计算loss
l.backward() # 做完后计算图会释放
print('\tgrad:', x, y, w.grad.item())# item取值，要是张量计算图一直累积
w.data -= 0.01 * w.grad.data# 不取data会是TENSOR有计算图w.grad.data.zero_()# 计算出来的梯度不清零会累加
print("progress:", epoch, l.item())
print('predict (after training)', 4, forward(4).item())
Pytorch实战--线性回归
# 1、算预测值
# 2、算loss
# 3、梯度设为0，并反向传播
# 3、梯度更新import torchx_data = https://www.it610.com/article/torch.Tensor([[1.0], [2.0], [3.0]])
y_data = torch.Tensor([[2.0], [4.0], [6.0]])# 构造线性模型,后面都是使用这样的模板
# 至少实现两个函数，__init__构造函数和forward()前馈函数
# backward()会根据我们的计算图自动构建
# 可以继承Functions来构建自己的计算块
class LinerModel(torch.nn.Module):
def __init__(self):
# 调用父类的构造
super(LinerModel, self).__init__()
# 构造Linear这个对象，对输入数据做线性变换
# class torch.nn.Linear(in_features, out_features, bias=True)
# in_features - 每个输入样本的大小
# out_features - 每个输出样本的大小
# bias - 若设置为False，这层不会学习偏置。默认值：True
self.linear = torch.nn.Linear(1, 1)def forward(self, x):
y_pred = self.linear(x)
return y_predmodel = LinerModel() # 实例化，可调用
# 定义MSE(均方差)损失函数，size_average=False不求均值
criterion = torch.nn.MSELoss(size_average=False)
# optim优化模块的SGD，第一个参数就是传递权重，model.parameters()model的所有权重
# 优化器对象
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)for epoch in range(100):
y_pred = model(x_data)
loss = criterion(y_pred, y_data)
# loss为一个对象，loss不会产生计算图，但会自动调用__str__()所以不会出错
print(epoch, loss)
# 梯度归零
optimizer.zero_grad()
# 反向传播
loss.backward()
# 根据梯度和预先设置的学习率进行更新（权重更新）
optimizer.step()# 打印权重和偏置值,weight是一个值但是一个矩阵
print('w=', model.linear.weight.item())
print('b=', model.linear.bias.item())# 测试
x_test = torch.Tensor([4.0])
y_test = model(x_test)
print('y_pred=', y_test.data)结果图：
文章图片
Pytorch实战--逻辑回归
import torch
import torch.nn.functional as F
import numpy as np
import matplotlib.pyplot as plt
x_data = https://www.it610.com/article/torch.Tensor([[1.0], [2.0], [3.0]])
y_data = torch.Tensor([[0], [0], [1]])
##
class LogisticRegressionModel(torch.nn.Module):
def __init__(self): #构造函数
super(LogisticRegressionModel, self).__init__()
self.linear = torch.nn.Linear(1, 1) #线性层
def forward(self, x):
y_pred = F.sigmoid(self.linear(x)) #激活函数
return y_pred
model = LogisticRegressionModel()
##
criterion = torch.nn.BCELoss(size_average = False) #计算损失
optimizer = torch.optim.SGD(model.parameters(), lr = 0.01) #优化器
##
for epoch in range(1000):
y_pred = model(x_data)
loss = criterion(y_pred, y_data)
print(epoch, loss.item())
optimizer.zero_grad() # 梯度置0
loss.backward() # 计算梯度，反向传播
optimizer.step() # 更新参数
##
x = np.linspace(0, 10, 200)
x_t = torch.Tensor(x).view((200, 1))
y_t = model(x_t)
y = y_t.data.numpy()
plt.plot(x, y)
plt.plot([0, 10], [0.5, 0.5], c='r')
plt.xlabel('Hours')
plt.ylabel('Probability of Pass')
plt.grid()
plt.show()处理多维特征的输入
import numpy as np
import torchxy = np.loadtxt('diabetes.csv.gz', delimiter=',', dtype=np.float32)
x_data = https://www.it610.com/article/torch.from_numpy(xy[:, :-1])
# [-1]加中括号拿出来是矩阵，不加是向量
y_data = torch.from_numpy(xy[:, [-1]])class Model(torch.nn.Module):
def __init__(self):
super(Model, self).__init__()
self.linear1 = torch.nn.Linear(8, 6)
self.linear2 = torch.nn.Linear(6, 4)
self.linear3 = torch.nn.Linear(4, 1)
# 这是nn下的Sigmoid是一个模块没有参数，在function调用的Sigmoid是函数
self.sigmoid = torch.nn.Sigmoid()def forward(self, x):
x = self.sigmoid(self.linear1(x))
x = self.sigmoid(self.linear2(x))
x = self.sigmoid(self.linear3(x))
return xmodel = Model()
criterion = torch.nn.BCELoss(size_average=True)# 损失函数
optimizer = torch.optim.SGD(model.parameters(), lr=0.1)# 优化函数，随机梯度递减for epoch in range(100):
# 前馈
y_pred = model(x_data)
loss = criterion(y_pred, y_data)
print(epoch, loss.item())# 反馈
optimizer.zero_grad()
loss.backward()# 更新
optimizer.step()
加载数据集
import numpy as np
import torch
from torch.utils.data import Dataset# Dataset是一个抽象类，只能被继承，不能实例化
from torch.utils.data import DataLoader# 可以直接实例化'''
四步：准备数据集-设计模型-构建损失函数和优化器-周期训练
'''class DiabetesDataset(Dataset):
def __init__(self, filepath):
xy = np.loadtxt(filepath, delimiter=',', dtype=np.float32)
self.len = xy.shape[0]
self.x_data = https://www.it610.com/article/torch.from_numpy(xy[:, :-1])
self.y_data = torch.from_numpy(xy[:, [-1]])def __getitem__(self, index):# 实例化对象后，该类能支持下标操作，通过index拿出数据
return self.x_data[index], self.y_data[index]def __len__(self):
return self.lendataset = DiabetesDataset('diabetes.csv.gz')
# dataset数据集，batch_size小批量的容量，shuffle是否要打乱，num_workers要几个并行进程来读
# DataLoader的实例化对象不能直接使用，因为windows和linux的多线程运行不一样，所以一般要放在函数里运行
train_loader = DataLoader(dataset=dataset, batch_size=32, shuffle=True, num_workers=2)class Model(torch.nn.Module):
def __init__(self):
super(Model, self).__init__()
self.linear1 = torch.nn.Linear(8, 6)
self.linear2 = torch.nn.Linear(6, 4)
self.linear3 = torch.nn.Linear(4, 1)
# 这是nn下的Sigmoid是一个模块没有参数，在function调用的Sigmoid是函数
self.sigmoid = torch.nn.Sigmoid()def forward(self, x):
x = self.sigmoid(self.linear1(x))
x = self.sigmoid(self.linear2(x))
x = self.sigmoid(self.linear3(x))
return xmodel = Model()
criterion = torch.nn.BCELoss(size_average=True)# 损失函数
optimizer = torch.optim.SGD(model.parameters(), lr=0.1)# 优化函数，随机梯度递减# 变成嵌套循环，实现Mini-Batch
for epoch in range(100):
# 从数据集0开始迭代
# 可以简写为for i, (inputs, labels) in enumerate(train_loader, 0):
for i, data in enumerate(train_loader, 0):
# 准备数据
inputs, labels = data
# 前馈
y_pred = model(inputs)
loss = criterion(y_pred, labels)
print(epoch, i, loss.item())
# 反馈
optimizer.zero_grad()
loss.backward()
# 更新
optimizer.step()多分类问题
import torch
from torchvision import transforms# 对图像进行处理的工具
from torchvision import datasets
from torch.utils.data import DataLoader
import torch.nn.functional as F# 使用激活函数relu()的包
import torch.optim as optim# 优化器的包batch_size = 64
# 对图像进行预处理，将图像转换为
transform = transforms.Compose([
# 将原始图像PIL变为张量tensor(H*W*C),再将[0,255]区间转换为[0.1,1.0]
transforms.ToTensor(),
# 使用均值和标准差对张量图像进行归一化
transforms.Normalize((0.1307,), (0.3081,))
])train_dataset = datasets.MNIST(root='dataset/mnist/', train=True, download=True, transform=transform)
train_loader = DataLoader(train_dataset, shuffle=True, batch_size=batch_size)test_dataset = datasets.MNIST(root='dataset/mnist/', train=False, download=True, transform=transform)
test_loader = DataLoader(test_dataset, shuffle=False, batch_size=batch_size)class Net(torch.nn.Module):
def __init__(self):
super(Net, self).__init__()
self.l1 = torch.nn.Linear(784, 512)
self.l2 = torch.nn.Linear(512, 256)
self.l3 = torch.nn.Linear(256, 128)
self.l4 = torch.nn.Linear(128, 64)
self.l5 = torch.nn.Linear(64, 10)def forward(self, x):
# 改变形状，相当于numpy的reshape
# view中一个参数定为-1，代表动态调整这个维度上的元素个数，以保证元素的总数不变。
x = x.view(-1, 784)
x = F.relu(self.l1(x))
x = F.relu(self.l2(x))
x = F.relu(self.l3(x))
x = F.relu(self.l4(x))
return self.l5(x)model = Net()
# 交叉熵损失函数
criterion = torch.nn.CrossEntropyLoss()
# model.parameters()直接使用的模型的所有参数
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.5)# momentum动量def train(epoch):
running_loss = 0.0
# 返回了数据下标和数据
for batch_idx, data in enumerate(train_loader, 0):
# 送入两个张量，一个张量是64个图像的特征，一个张量图片对应的数字
inputs, target = data
# 梯度归零
optimizer.zero_grad()# forward+backward+update
outputs = model(inputs)
# 计算损失，用的交叉熵损失函数
loss = criterion(outputs, target)
# 反馈
loss.backward()
# 随机梯度下降更新
optimizer.step()# 每300次输出一次
running_loss += loss.item()
if batch_idx % 300 == 299:
print('[%d,%5d] loss:%.3f' % (epoch + 1, batch_idx + 1, running_loss / 300))
running_loss = 0.0def test():
correct = 0
total = 0
# 不会计算梯度
with torch.no_grad():
for data in test_loader:# 拿数据
images, labels = data
outputs = model(images)# 预测
# outputs.data是一个矩阵，每一行10个量，最大值的下标就是预测值
_, predicted = torch.max(outputs.data, dim=1)# 沿着第一维度，找最大值的下标，返回最大值和下标
total += labels.size(0)# labels.size(0)=64 每个都是64个元素，就可以计算总的元素
# (predicted == labels).sum()这个是张量，而加了item()变为一个数字，即相等的数量
correct += (predicted == labels).sum().item()
print('Accuracy on test set:%d %%' % (100 * correct / total))# 正确的数量除以总数if __name__ == '__main__':
for epoch in range(10):
train(epoch)
test()卷积神经网络
简单的构建
import torch# 输入的通道就是上图的n,输出的通道就是上图的m
in_channels, out_channels = 5, 10
width, height = 100, 100# 图像的大小
kernel_size = 3# 卷积盒的大小
batch_size = 1# 批量大小# 随机生成了一个小批量=1的5*100*100的张量
input = torch.randn(batch_size, in_channels, width, height)# Conv2d对由多个输入平面组成的输入信号进行二维卷积
conv_layer = torch.nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size)output = conv_layer(input)# print(input)
print(input.shape)
print(output.shape)
print(conv_layer.weight.shape)
运行结果：
torch.Size([1, 5, 100, 100])
torch.Size([1, 10, 98, 98])
torch.Size([10, 5, 3, 3])
paddingimport torchinput = [3, 4, 6, 5, 7,
2, 4, 6, 8, 2,
1, 6, 7, 8, 4,
9, 7, 4, 6, 2,
3, 7, 5, 4, 1]input = torch.Tensor(input).view(1, 1, 5, 5)# bias=False不加偏置量
conv_layer = torch.nn.Conv2d(1, 1, kernel_size=3, padding=1, bias=False)kernel = torch.Tensor([1, 2, 3, 4, 5, 6, 7, 8, 9]).view(1, 1, 3, 3)
# 把kernel赋值给卷积层权重，做初始化
conv_layer.weight.data = https://www.it610.com/article/kernel.dataoutput = conv_layer(input)
print(output)
运行结果：
tensor([[[[ 91., 168., 224., 215., 127.],
[114., 211., 295., 262., 149.],
[192., 259., 282., 214., 122.],
[194., 251., 253., 169.,86.],
[ 96., 112., 110.,68.,31.]]]], grad_fn=)
Layer-stride步长import torchinput = [3, 4, 6, 5, 7,
2, 4, 6, 8, 2,
1, 6, 7, 8, 4,
9, 7, 4, 6, 2,
3, 7, 5, 4, 1]input = torch.Tensor(input).view(1, 1, 5, 5)# bias=False不加偏置量
conv_layer = torch.nn.Conv2d(1, 1, kernel_size=3, stride=2, bias=False)kernel = torch.Tensor([1, 2, 3, 4, 5, 6, 7, 8, 9]).view(1, 1, 3, 3)
# 把kernel赋值给卷积层权重，做初始化
conv_layer.weight.data = https://www.it610.com/article/kernel.dataoutput = conv_layer(input)
print(output)运行结果：
tensor([[[[211., 262.],
[251., 169.]]]], grad_fn=)
Max Pooling Layer最大池化层
最大池化层是没有权重的
import torchinput = [3, 9, 6, 5,
2, 4, 6, 8,
1, 6, 2, 1,
3, 7, 4, 6]input = torch.Tensor(input).view(1, 1, 4, 4)maxpooling_layer = torch.nn.MaxPool2d(kernel_size=2)output = maxpooling_layer(input)
print(output)运行结果：
tensor([[[[9., 8.],
[7., 6.]]]])
import torch
from torchvision import transforms# 对图像进行处理的工具
from torchvision import datasets
from torch.utils.data import DataLoader
import torch.nn.functional as F# 使用激活函数relu()的包
import torch.optim as optim# 优化器的包batch_size = 64
# 对图像进行预处理，将图像转换为
transform = transforms.Compose([
# 将原始图像PIL变为张量tensor(H*W*C),再将[0,255]区间转换为[0.1,1.0]
transforms.ToTensor(),
# 使用均值和标准差对张量图像进行归一化
transforms.Normalize((0.1307,), (0.3081,))
])train_dataset = datasets.MNIST(root='dataset/mnist/', train=True, download=True, transform=transform)
train_loader = DataLoader(train_dataset, shuffle=True, batch_size=batch_size)test_dataset = datasets.MNIST(root='dataset/mnist/', train=False, download=True, transform=transform)
test_loader = DataLoader(test_dataset, shuffle=False, batch_size=batch_size)class Net(torch.nn.Module):
def __init__(self):
super(Net, self).__init__()
# 定义两个卷积层
self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=5)
self.conv2 = torch.nn.Conv2d(10, 20, kernel_size=5)
# 定义一个池化层
self.pooling = torch.nn.MaxPool2d(2)
# 定义一个全连接的线性层
self.fc = torch.nn.Linear(320, 10)def forward(self, x):
# Flatten data from (n, 1, 28, 28) to (n, 784)
# x.size(0)就是取的n
batch_size = x.size(0)
# 用relu做非线性激活
# 先做卷积再做池化再做relu
x = F.relu(self.pooling(self.conv1(x)))
x = F.relu(self.pooling(self.conv2(x)))
# 做view把数据变为做全连接网络所需要的输入
x = x.view(batch_size, -1)
return self.fc(x)
# 因为最后一层要做交叉熵损失，所以最后一层不做激活model = Net()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model.to(device)
# 交叉熵损失函数
criterion = torch.nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.5)# momentum动量def train(epoch):
running_loss = 0.0
# 返回了数据下标和数据
for batch_idx, data in enumerate(train_loader, 0):
# 送入两个张量，一个张量是64个图像的特征，一个张量图片对应的数字
inputs, target = data
# 把输入输出迁入GPU
inputs, target = inputs.to(device), target.to(device)
# 梯度归零
optimizer.zero_grad()# forward+backward+update
outputs = model(inputs)
# 计算损失，用的交叉熵损失函数
loss = criterion(outputs, target)
# 反馈
loss.backward()
# 随机梯度下降更新
optimizer.step()# 每300次输出一次
running_loss += loss.item()
if batch_idx % 300 == 299:
print('[%d,%5d] loss:%.3f' % (epoch + 1, batch_idx + 1, running_loss / 300))
running_loss = 0.0def test():
correct = 0
total = 0
# 不会计算梯度
with torch.no_grad():
for data in test_loader:# 拿数据
images, labels = data
images, labels = images.to(device), labels.to(device)
outputs = model(images)# 预测
# outputs.data是一个矩阵，每一行10个量，最大值的下标就是预测值
_, predicted = torch.max(outputs.data, dim=1)# 沿着第一维度，找最大值的下标，返回最大值和下标
total += labels.size(0)# labels.size(0)=64 每个都是64个元素，就可以计算总的元素
# (predicted == labels).sum()这个是张量，而加了item()变为一个数字，即相等的数量
correct += (predicted == labels).sum().item()
print('Accuracy on test set:%d %%' % (100 * correct / total))# 正确的数量除以总数if __name__ == '__main__':
for epoch in range(10):
train(epoch)
test()卷积神经网络（高级）
import torch
import torch.nn as nn
from torchvision import transforms# 对图像进行处理的工具
from torchvision import datasets
from torch.utils.data import DataLoader
import torch.nn.functional as F# 使用激活函数relu()的包
import torch.optim as optim# 优化器的包batch_size = 64
# 对图像进行预处理，将图像转换为
transform = transforms.Compose([
# 将原始图像PIL变为张量tensor(H*W*C),再将[0,255]区间转换为[0.1,1.0]
transforms.ToTensor(),
# 使用均值和标准差对张量图像进行归一化
transforms.Normalize((0.1307,), (0.3081,))
])train_dataset = datasets.MNIST(root='dataset/mnist/', train=True, download=True, transform=transform)
train_loader = DataLoader(train_dataset, shuffle=True, batch_size=batch_size)test_dataset = datasets.MNIST(root='dataset/mnist/', train=False, download=True, transform=transform)
test_loader = DataLoader(test_dataset, shuffle=False, batch_size=batch_size)class InceptionA(nn.Module):
def __init__(self, in_channels):
super(InceptionA, self).__init__()
# 第一个通道，输入通道为in_channels,输出通道为16，卷积盒的大小为1*1的卷积层
self.branch1x1 = nn.Conv2d(in_channels, 16, kernel_size=1)# 第二个通道
self.branch5x5_1 = nn.Conv2d(in_channels, 16, kernel_size=1)
self.branch5x5_2 = nn.Conv2d(16, 24, kernel_size=5, padding=2)# 第三个通道
self.branch3x3_1 = nn.Conv2d(in_channels, 16, kernel_size=1)
self.branch3x3_2 = nn.Conv2d(16, 24, kernel_size=3, padding=1)
self.branch3x3_3 = nn.Conv2d(24, 24, kernel_size=3, padding=1)# 第四个通道
self.branch_pool = nn.Conv2d(in_channels, 24, kernel_size=1)def forward(self, x):
branch1x1 = self.branch1x1(x)branch5x5 = self.branch5x5_1(x)
branch5x5 = self.branch5x5_2(branch5x5)branch3x3 = self.branch3x3_1(x)
branch3x3 = self.branch3x3_2(branch3x3)
branch3x3 = self.branch3x3_3(branch3x3)branch_pool = F.avg_pool2d(x, kernel_size=3, stride=1, padding=1)
branch_pool = self.branch_pool(branch_pool)# 拼接
outputs = [branch1x1, branch5x5, branch3x3, branch_pool]
return torch.cat(outputs, dim=1)class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
self.conv2 = nn.Conv2d(88, 20, kernel_size=5)self.incep1 = InceptionA(in_channels=10)
self.incep2 = InceptionA(in_channels=20)self.mp = nn.MaxPool2d(2)
self.fc = nn.Linear(1408, 10)def forward(self, x):
in_size = x.size(0)
x = F.relu(self.mp(self.conv1(x)))
x = self.incep1(x)
x = F.relu(self.mp(self.conv2(x)))
x = self.incep2(x)
x = x.view(in_size, -1)
x = self.fc(x)
return xmodel = Net()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model.to(device)
# 交叉熵损失函数
criterion = torch.nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.5)# momentum动量def train(epoch):
running_loss = 0.0
# 返回了数据下标和数据
for batch_idx, data in enumerate(train_loader, 0):
# 送入两个张量，一个张量是64个图像的特征，一个张量图片对应的数字
inputs, target = data
# 把输入输出迁入GPU
inputs, target = inputs.to(device), target.to(device)
# 梯度归零
optimizer.zero_grad()# forward+backward+update
outputs = model(inputs)
# 计算损失，用的交叉熵损失函数
loss = criterion(outputs, target)
# 反馈
loss.backward()
# 随机梯度下降更新
optimizer.step()# 每300次输出一次
running_loss += loss.item()
if batch_idx % 300 == 299:
print('[%d,%5d] loss:%.3f' % (epoch + 1, batch_idx + 1, running_loss / 300))
running_loss = 0.0def test():
correct = 0
total = 0
# 不会计算梯度
with torch.no_grad():
for data in test_loader:# 拿数据
images, labels = data
images, labels = images.to(device), labels.to(device)
outputs = model(images)# 预测
# outputs.data是一个矩阵，每一行10个量，最大值的下标就是预测值
_, predicted = torch.max(outputs.data, dim=1)# 沿着第一维度，找最大值的下标，返回最大值和下标
total += labels.size(0)# labels.size(0)=64 每个都是64个元素，就可以计算总的元素
# (predicted == labels).sum()这个是张量，而加了item()变为一个数字，即相等的数量
correct += (predicted == labels).sum().item()
print('Accuracy on test set:%d %%' % (100 * correct / total))# 正确的数量除以总数if __name__ == '__main__':
for epoch in range(10):
train(epoch)
test()Residual net残差结构块
定义的该层输入和输出的大小是一样的
import torch.nn as nn
import torch.nn.functional as Fclass ResidualBlock(nn.Module):
def __init__(self,channels):
super(ResidualBlock,self).__init__()
self.channels = channels
self.conv1 = nn.Conv2d(channels,channels,kernel_size=3,padding=1)
self.conv2 = nn.Conv2d(channels,channels,kernel_size=3,padding=1)def forward(self,x):
y = F.relu(self.conv1(x))
y = self.conv2(y)
return F.relu(x+y)class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(1, 16, kernel_size=5)
self.conv2 = nn.Conv2d(16, 32, kernel_size=5)
self.mp = nn.MaxPool2d(2)self.rblock1 = ResidualBlock(16)
self.rblock2 = ResidualBlock(32)self.fc = nn.Linear(512, 10)def forward(self, x):
in_size = x.size(0)
x = self.mp(F.relu(self.conv1(x)))
x = self.rblock1(x)
x = self.mp(F.relu(self.conv2(x)))
x = self.rblock2(x)
x = x.view(in_size, -1)
x = self.fc(x)
return x
RNN
1.准备数据
定义一个数据集类，并读取数据文件。
from torch.utils.data import Dataset
import pandas as pdclass NameDataset(Dataset):
"""数据集类"""
def __init__(self, is_train_set=True):
filename = './name_data/names_train.csv' if is_train_set else './name_data/names_test.csv'
data = https://www.it610.com/article/pd.read_csv(filename, header=None)
self.names = data[0]
self.len = len(self.names)
self.countries = data[1]
self.country_list = list(sorted(set(self.countries)))
self.country_dict = self.getCountryDict()
self.country_num = len(self.country_list)def __getitem__(self, index):
return self.names[index], self.country_dict[self.countries[index]]def __len__(self):
return self.lendef idx2country(self, index):
return self.country_list[index]def getCountryDict(self):
country_dict = dict()
for idx, country_name in enumerate(self.country_list, 0):
country_dict[country_name] = idx
return country_dictdef getCountriesNum(self):
return self.country_num
定义函数，用于将读取到的数据转化为tensor。
def name2list(name):
"""返回ASCII码表示的姓名列表与列表长度"""
arr = [ord(c) for c in name]
return arr, len(arr)def make_tensors(names, countries):
# 元组列表，每个元组包含ASCII码表示的姓名列表与列表长度
sequences_and_lengths = [name2list(name) for name in names]
# 取出所有的ASCII码表示的姓名列表
name_sequences = [sl[0] for sl in sequences_and_lengths]
# 取出所有的列表长度
seq_lengths = torch.LongTensor([sl[1] for sl in sequences_and_lengths])
# 将countries转为long型
countries = countries.long()# 接下来每个名字序列补零，使之长度一样。
# 先初始化一个全为零的tensor，大小为 所有姓名的数量*最长姓名的长度
seq_tensor = torch.zeros(len(name_sequences), seq_lengths.max()).long()# 将姓名序列覆盖到初始化的全零tensor上
for idx, (seq, seq_len) in enumerate(zip(name_sequences, seq_lengths), 0):
seq_tensor[idx, :seq_len] = torch.LongTensor(seq)
# 根据序列长度seq_lengths对补零后tensor进行降序怕排列，方便后面加速计算。
# 返回排序后的seq_lengths与索引变化列表
seq_lengths, perm_idx = seq_lengths.sort(dim=0, descending=True)
# 根据索引变化列表对ASCII码表示的姓名列表进行排序
seq_tensor = seq_tensor[perm_idx]
# 根据索引变化列表对countries进行排序，使姓名与国家还是一一对应关系
# seq_tensor.shape : batch_size*max_seq_lengths,
# seq_lengths.shape : batch_size
# countries.shape : batch_size
countries = countries[perm_idx]
return seq_tensor, seq_lengths, countries
2.定义模型
import torch
from torch.nn.utils.rnn import pack_padded_sequenceclass RNNClassifier(torch.nn.Module):
# input_size=128, hidden_size=100, output_size=18
def __init__(self, input_size, hidden_size, output_size, n_layers=1, bidirectional=True):
super(RNNClassifier, self).__init__()
self.hidden_size = hidden_size
self.n_layers = n_layers
self.n_directions = 2 if bidirectional else 1# 是否双向
self.embedding = torch.nn.Embedding(input_size, hidden_size)# 输入大小128，输出大小100。
# 经过Embedding后input的大小是100，hidden_size的大小也是100，所以形参都是hidden_size。
self.gru = torch.nn.GRU(hidden_size, hidden_size, n_layers, bidirectional=bidirectional)
# 如果是双向，会输出两个hidden层，要进行拼接，所以线形成的input大小是 hidden_size * self.n_directions，输出是大小是18，是为18个国家的概率。
self.fc = torch.nn.Linear(hidden_size * self.n_directions, output_size)def _init_hidden(self, batch_size):
hidden = torch.zeros(self.n_layers * self.n_directions, batch_size, self.hidden_size)
return hiddendef forward(self, input, seq_lengths):
# 先对input进行转置，input shape : batch_size*max_seq_lengths -> max_seq_lengths*batch_size 每一列表示姓名
input = input.t()
batch_size = input.size(1)# 总共有多少列，既是batch_size的大小
hidden = self._init_hidden(batch_size)# 初始化隐藏层
embedding = self.embedding(input)# embedding.shape : max_seq_lengths*batch_size*hidden_size 12*64*100
# pack_padded_sequence方便批量计算
gru_input = pack_padded_sequence(embedding, seq_lengths)
# 进入网络进行计算
output, hidden = self.gru(gru_input, hidden)# 如果是双向的，需要进行拼接
if self.n_directions == 2:
hidden_cat = torch.cat([hidden[-1], hidden[-2]], dim=1)else:
hidden_cat = hidden[-1]# 线性层输出大小为18
fc_output = self.fc(hidden_cat)
return fc_output
3.定义训练函数
def time_since(since):
s = time.time() - since
m = math.floor(s/60)
s-= m*60
return '%dm %ds' % (m, s)def trainModel():
total_loss = 0
for i, (names, countries) in enumerate(trainloader, 1):# 这里的1意思是 i 从1开始。
# make_tensors函数返回经过降序排列后的 姓名列表，列表长度，国家
inputs, seq_lengths, target = make_tensors(names, countries)
# 输入姓名列表与列表长度向前计算
output = classifier(inputs, seq_lengths)
loss = criterion(output, target)
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_loss += loss.item()
if i % 10 == 0:
print(i)
print(f'[{time_since(start)}] Epoch {epoch} ', end='')
print(f'[{i * len(inputs)}/{len(trainset)}] ', end='')
print(f'loss={total_loss / (i * len(inputs))}')
return total_loss
4.定义测试函数，跟训练函数相差不大
def testModel():
correct = 0
total = len(testset)
print("evaluating trained model ...")
with torch.no_grad():
for i, (names, countries) in enumerate(testloader, 1):
inputs, seq_lengths, target = make_tensors(names, countries)
output = classifier(inputs, seq_lengths)
pred = output.max(dim=1, keepdim=True)[1]
correct += pred.eq(target.view_as(pred)).sum().item()
percent = '%.2f' % (100 * correct / total)
print(f'Test set: Accuracy {correct}/{total} {percent}%')
return correct / total
5.主函数循环
from torch.utils.data import DataLoader
import time
import mathif __name__ == '__main__':N_EPOCHS = 30# epoch
HIDDEN_SIZE = 100# 隐藏层的大小，也是Embedding后输出的大小
BATCH_SIZE = 64
N_COUNTRY = 18# 总共有18个类别的国家，为RNN后输出的大小
N_LAYER = 2
N_CHARS = 128# 字母字典的大小，Embedding输入的大小trainset = NameDataset(is_train_set=True)
trainloader = DataLoader(trainset, batch_size=BATCH_SIZE, shuffle=True)testset = NameDataset(is_train_set=False)
testloader = DataLoader(testset, batch_size=BATCH_SIZE, shuffle=False)# 建立分类模型
classifier = RNNClassifier(N_CHARS, HIDDEN_SIZE, N_COUNTRY, N_LAYER)# 建立损失函数与优化器
criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(classifier.parameters(), lr=0.001)start = time.time()
print("Training for %d epochs..." % N_EPOCHS)
acc_list = []
for epoch in range(1, N_EPOCHS + 1):
# Train cycle
trainModel()
acc = testModel()
acc_list.append(acc)推荐阅读
- 深度学习|PyTorch学习笔记（四）（PyTorch基础实战）
- 深度学习|PyTorch学习笔记（三）（PyTorch主要组成模块）
- Pytorch|PyTorch学习笔记（二）（简介与基础知识）
- 深度学习|深度学习 --- 卷积神经网络CNN（LeNet-5网络详解）
- 计算机视觉|OpenCV图像轮廓（2）轮廓面积
- 深度学习|YOLOv1——v5学习笔记
- 计算机视觉|OpenCV图像轮廓（1）查找和绘制轮廓
- 僧旅|神经网络模型LeNet-5学习
- 神经网络|【机器学习基础】常用激活函数（激励函数）理解与总结