项目结构
项目结构如图,代码都放在mnistclassify.py里面,data数据是代码执行过程中自己下载的。
项目代码
- 导入包,构建训练集测试集
from random import shuffle
from turtle import forward
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torchvision import datasets,transforms
import matplotlib.pyplot as plt
import numpy as np
# 定义超参数
input_size = 28
num_classes = 10
num_epoches = 3
batch_size = 64
# 训练集
train_dateset = datasets.MNIST(root='./data',train=True,transform=transforms.ToTensor(),download=True)
# 测试集
test_dateset = datasets.MNIST(root='./data',train=True,transform=transforms.ToTensor())
# 构建batch数据
train_loader = torch.utils.data.DataLoader(dataset=train_dateset,batch_size=batch_size,shuffle=True)
test_loader = torch.utils.data.DataLoader(dataset=test_dateset,batch_size=batch_size,shuffle=True)
- 构建神经网络
# 构建网络
class CNN(nn.Module):
def __init__(self) -> None:
super(CNN, self).__init__()
self.conv1 = nn.Sequential(
nn.Conv2d(
in_channels=1, # 灰度图
out_channels=16, # 输出特征图个数
kernel_size=5, # 卷积核大小
stride=1, # 步长
padding=2, # 边缘填充,如果stride=1,希望卷积后的图像和原来的图像一样大则设置padding=(kernal_size-1)/2
), # 输出特征图为(16,28,28)
nn.ReLU(),
nn.MaxPool2d(kernel_size=2) # 2*2最大池化,结果为(16,14,14)
)
self.conv2 = nn.Sequential( # 输入(16,14,14)
nn.Conv2d(16, 32, 5, 1, 2), # 输出(32,14,14)
nn.ReLU(),
nn.MaxPool2d(2), # 输出(32,7,7)
)
self.out = nn.Linear(32 * 7 *7, 10) # 全连接得到结果
def forward(self, x):
x = self.conv1(x)
x = self.conv2(x)
x = x.view(x.size(0), -1) # 将结果转换为向量,方便下一步全连接(32*7*7)
output = self.out(x)
return output
- 实例化网络开始训练
# 预测准确率
def accuracy(predictins, labels):
pred = torch.max(predictins.data, 1)[1]
rights = pred.eq(labels.data.view_as(pred)).sum()
return rights, len(labels)
# 实例化神经网络
net = CNN()
# 损失函数
criterion = nn.CrossEntropyLoss()
# 优化器
optimizer = optim.Adam(net.parameters(), lr=0.001)
# 开始训练循环
for epoch in range(num_epoches):
# 保存当前epoch结果
train_rights = []
for batch_idx, (data, target) in enumerate(train_loader):
net.train()
output = net(data)
loss = criterion(output, target)
optimizer.zero_grad()
loss.backward()
optimizer.step()
right = accuracy(output, target)
train_rights.append(right)
if batch_idx % 100 == 0:
net.eval()
val_rights = []
for (data, target) in test_loader:
output = net(data)
right = accuracy(output, target)
val_rights.append(right)
# 准确率计算
train_r = (sum([tup[0] for tup in train_rights]), sum([tup[1] for tup in train_rights]))
val_r = (sum([tup[0] for tup in val_rights]), sum([tup[1] for tup in val_rights]))
print('当前epoch:{} [{}/{}({:.0f}%)]\t损失: {:.6f}\t训练集准确率: {:.2f}%\t测试集准确率: {:.2f}%'.format(
epoch, batch_idx * batch_size, len(train_loader.dataset),
100. * batch_idx / len(train_loader),
loss.data,
100. * train_r[0].numpy() / train_r[1],
100. * val_r[0].numpy() / val_r[1],
))
- 训练结果
当前epoch:0 [0/60000(0%)] 损失: 2.290263 训练集准确率: 6.25% 测试集准确率: 11.39%
当前epoch:0 [6400/60000(11%)] 损失: 0.222888 训练集准确率: 76.14% 测试集准确率: 90.28%
当前epoch:0 [12800/60000(21%)] 损失: 0.275965 训练集准确率: 84.60% 测试集准确率: 94.70%
当前epoch:0 [19200/60000(32%)] 损失: 0.071834 训练集准确率: 88.24% 测试集准确率: 95.60%
当前epoch:0 [25600/60000(43%)] 损失: 0.029019 训练集准确率: 90.25% 测试集准确率: 96.68%
当前epoch:0 [32000/60000(53%)] 损失: 0.159890 训练集准确率: 91.48% 测试集准确率: 97.08%
当前epoch:0 [38400/60000(64%)] 损失: 0.080257 训练集准确率: 92.39% 测试集准确率: 97.00%
当前epoch:0 [44800/60000(75%)] 损失: 0.100067 训练集准确率: 93.11% 测试集准确率: 97.57%
当前epoch:0 [51200/60000(85%)] 损失: 0.105826 训练集准确率: 93.66% 测试集准确率: 97.84%
当前epoch:0 [57600/60000(96%)] 损失: 0.042444 训练集准确率: 94.11% 测试集准确率: 98.05%
当前epoch:1 [0/60000(0%)] 损失: 0.169493 训练集准确率: 93.75% 测试集准确率: 98.01%
当前epoch:1 [6400/60000(11%)] 损失: 0.033878 训练集准确率: 98.04% 测试集准确率: 97.87%
当前epoch:1 [12800/60000(21%)] 损失: 0.108467 训练集准确率: 98.05% 测试集准确率: 98.01%
当前epoch:1 [19200/60000(32%)] 损失: 0.007603 训练集准确率: 97.97% 测试集准确率: 98.35%
当前epoch:1 [25600/60000(43%)] 损失: 0.202825 训练集准确率: 98.04% 测试集准确率: 98.49%
当前epoch:1 [32000/60000(53%)] 损失: 0.113783 训练集准确率: 98.11% 测试集准确率: 98.47%
当前epoch:1 [38400/60000(64%)] 损失: 0.027782 训练集准确率: 98.11% 测试集准确率: 98.46%
当前epoch:1 [44800/60000(75%)] 损失: 0.034398 训练集准确率: 98.12% 测试集准确率: 98.51%
当前epoch:1 [51200/60000(85%)] 损失: 0.013913 训练集准确率: 98.18% 测试集准确率: 98.51%
当前epoch:1 [57600/60000(96%)] 损失: 0.021681 训练集准确率: 98.19% 测试集准确率: 98.91%
当前epoch:2 [0/60000(0%)] 损失: 0.052889 训练集准确率: 96.88% 测试集准确率: 98.72%
当前epoch:2 [6400/60000(11%)] 损失: 0.070504 训练集准确率: 98.95% 测试集准确率: 98.86%
当前epoch:2 [12800/60000(21%)] 损失: 0.104337 训练集准确率: 98.67% 测试集准确率: 98.85%
当前epoch:2 [19200/60000(32%)] 损失: 0.028965 训练集准确率: 98.72% 测试集准确率: 98.70%
当前epoch:2 [25600/60000(43%)] 损失: 0.048499 训练集准确率: 98.70% 测试集准确率: 98.82%
当前epoch:2 [32000/60000(53%)] 损失: 0.021659 训练集准确率: 98.70% 测试集准确率: 98.80%
当前epoch:2 [38400/60000(64%)] 损失: 0.002921 训练集准确率: 98.72% 测试集准确率: 98.95%
当前epoch:2 [44800/60000(75%)] 损失: 0.015612 训练集准确率: 98.70% 测试集准确率: 98.92%
当前epoch:2 [51200/60000(85%)] 损失: 0.043291 训练集准确率: 98.71% 测试集准确率: 99.08%
当前epoch:2 [57600/60000(96%)] 损失: 0.033159 训练集准确率: 98.72% 测试集准确率: 99.01%
如有代码不懂或者报错请评论区留言,博主帮忙解答、调试。