当前位置:网站首页>pytorch tear CNN
pytorch tear CNN
2022-08-10 22:41:00 【Python ml】
import torch
from torch.utils.data import DataLoader # We want to load the data sets of
from torchvision import transforms # The raw data processing
from torchvision import datasets # pytorchVery thoughtful for us directly the data set
import torch.nn.functional as F # 激活函数
import torch.optim as optim
import matplotlib.pyplot as plt
batch_size = 64
# We get the picture ispillow,We want to convert him to training model cantensorThat is the format of the tensor
transform = transforms.Compose([transforms.ToTensor()])
# 加载训练集,pytorchVery thoughtful for us directly the data set,注意,Even if you don't have to download the data sets
# 数据下载
train_dataset = datasets.MNIST(root='E:/data/cnn', train=True, download=True, transform=transform)
# 数据打包
train_loader = DataLoader(dataset=train_dataset, shuffle=True, batch_size=batch_size)
# Same way to load the test set
test_dataset = datasets.MNIST(root='E:/data/cnn', train=False, download=True, transform=transform)
test_loader = DataLoader(dataset=test_dataset, shuffle=False, batch_size=batch_size)
# Next we look at the model is how to do
class Net(torch.nn.Module):
def __init__(self):
super(Net, self).__init__()
# Defines our first used convolution layer,Because image input channel for1,第一个参数就是1
# The output channel for10,kernel_size是卷积核的大小,这里定义的是5x5的
self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=5)
# Understand the definition of the above,The following are you sure you can understand
self.conv2 = torch.nn.Conv2d(10, 20, kernel_size=5)
# To define a pooling layer
self.pooling = torch.nn.MaxPool2d(2)
# Finally we did classification using linear layer
self.fc = torch.nn.Linear(320, 10)
# The following is the process of calculation
def forward(self, x):
# Flatten data from (n, 1, 28, 28) to (n, 784)
batch_size = x.size(0) # 这里面的0是xThe size of the first1个参数,自动获取batch大小
# 输入x经过一个卷积层,After a pooling layer,最后用relu做激活
x = F.relu(self.pooling(self.conv1(x)))
# Go through the above process
x = F.relu(self.pooling(self.conv2(x)))
# In order to give us the final linear layer of a fully connected with
# We are going to take a two-dimensional picture(Actually here is processed)20x4x4Tensor into a d
x = x.view(batch_size, -1) # flatten
# 经过线性层,Sure he is0~9The probability of each number
x = self.fc(x)
return x
model = Net() # 实例化模型
# Moved the computing andGPU
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model.to(device)
# 定义一个损失函数,We model to calculate the output value and standard value of the gap between
criterion = torch.nn.CrossEntropyLoss()
# 定义一个优化器,Training model zha training,就靠这个,He would reverse the changes to the corresponding layer weights
optimizer = optim.SGD(model.parameters(), lr=0.1, momentum=0.5) # lr为学习率
def train(epoch):
running_loss = 0.0
for batch_idx, data in enumerate(train_loader, 0): # 每次取一个样本
inputs, target = data
inputs, target = inputs.to(device), target.to(device)
# 优化器清零
optimizer.zero_grad()
# Forward to calculate the
outputs = model(inputs)
# 计算损失
loss = criterion(outputs, target)
# 反向求梯度
loss.backward()
# 更新权重
optimizer.step()
# Add up losses
running_loss += loss.item()
# 每300Output the data
if batch_idx % 300 == 299:
print('[%d, %5d] loss: %.3f' % (epoch + 1, batch_idx + 1, running_loss / 2000))
running_loss = 0.0
def CNN_test():
correct = 0
total = 0
with torch.no_grad(): # 不用算梯度
for data in test_loader:
inputs, target = data
inputs, target = inputs.to(device), target.to(device)
outputs = model(inputs)
# We take the highest probability that count as the output
_, predicted = torch.max(outputs.data, dim=1)
total += target.size(0)
# 计算正确率
correct += (predicted == target).sum().item()
print('Accuracy on test set: %d %% [%d/%d]' % (100 * correct / total, correct, total))
return correct / total
if __name__ == '__main__':
print("开始训练")
epoch_list = []
acc_list = []
for epoch in range(10):
print("epoch:%d" % epoch)
train(epoch)
acc = CNN_test()
epoch_list.append(epoch)
acc_list.append(acc)
plt.plot(epoch_list, acc_list)
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.show()
边栏推荐
- Research on multi-element N-k fault model of power system based on AC power flow (implemented by Matlab code) [Power System Fault]
- 合并k个已排序的链表
- IM 即时通讯开发如何设计图片文件的服务端存储架构
- 电力系统潮流计算(牛顿-拉夫逊法、高斯-赛德尔法、快速解耦法)(Matlab代码实现)
- How to secure users in LDAP directory service?
- "DevOps Night Talk" - Pilot - Introduction to CNCF Open Source DevOps Project DevStream - feat. PMC member Hu Tao
- JVM classic fifty questions, now the interview is stable
- 为什么一般公司面试结束后会说「回去等消息」,而不是直接告诉面试者结果?
- 什么是Jmeter?Jmeter使用的原理步骤是什么?
- QT笔记——vs + qt 创建一个带界面的 dll 和 调用带界面的dll
猜你喜欢
《DevOps围炉夜话》- Pilot - CNCF开源DevOps项目DevStream简介 - feat. PMC成员胡涛
罗克韦尔AB PLC RSLogix5000中计数器指令使用方法介绍
ThreadLocal comprehensive analysis (1)
合并k个已排序的链表
Use Cloudreve to build a private cloud disk
Common interview questions for APP UI automation testing, maybe useful~
解码2022中国网安强星丨正向建、反向查,华为构建数字化时代的网络安全防线
win系统下pytorch深度学习环境安装
Research on multi-element N-k fault model of power system based on AC power flow (implemented by Matlab code) [Power System Fault]
QT笔记——用VS + qt 生成dll 和 调用生成的dll
随机推荐
These must-know JVM knowledge, I have sorted it out with a mind map
port forwarding
CIKM2022 | 基于双向Transformers对比学习的序列推荐
QT笔记——vs + qt 创建一个带界面的 dll 和 调用带界面的dll
camera预览流程 --- 从HAL到OEM
OneNote 教程,如何在 OneNote 中整理笔记本?
瑞幸咖啡第二季营收33亿:门店达7195家 更换CFO
异常的了解
SDP
Shell 编程--Sed
H3C S5130 IRF做堆叠
3598. 二叉树遍历(华中科技大学考研机试题)
shell (text printing tool awk)
Shell编程之条件语句(二)
云服务器基于 SSH 协议实现免密登录
一篇文章教你Pytest快速入门和基础讲解,一定要看
pytorch手撕CNN
2021IDEA创建web工程
B站数据分析岗实习生面试记录
unusual understanding