1.Abstract

2.Background

3.Task

手写识别和评估性能

4.Work

手写

import torch
import torchvision
from torchvision import datasets, transforms
# 1. 加载MNIST手写数字数据集数据和标签
transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.5, ), (0.5, ))])
trainset = datasets.MNIST(root='./data', train=True,
                            download=True, transform=transform)
trainsetloader = torch.utils.data.DataLoader(trainset, batch_size=20000, shuffle=True)

testset = datasets.MNIST(root='./data', train=True,
                            download=True, transform=transform)
testsetloader = torch.utils.data.DataLoader(testset, batch_size=20000, shuffle=True)

#######如果你不放心数据有没有加载出可以将图片显示出来看下#######
# dataiter = iter(trainsetloader)
# images, labels = dataiter.next()
# import numpy as np
# import matplotlib.pyplot as plt
# plt.imshow(images[0].numpy().squeeze())
# plt.show()
# print(images.shape)
# print(labels.shape)
##########上面这段是显示图片的代码#############


# 2. 设计网络结构
first_in, first_out, second_out = 28*28,  128, 10
model = torch.nn.Sequential(
    torch.nn.Linear(first_in, first_out),
    torch.nn.ReLU(),
    torch.nn.Linear(first_out, second_out),
)

# 3. 设计损失函数
loss_fn = torch.nn.CrossEntropyLoss()

# 4. 设置用于自动调节神经网络参数的优化器
learning_rate = 1e-4
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)

# 5. 训练神经网络(重复训练10次)
for t in range(10):
    for i, one_batch in enumerate(trainsetloader,0):
        data,label = one_batch
        data[0].view(1,784)# 将28x28的图片变成784的向量
        data = data.view(data.shape[0],-1)

        # 让神经网络根据现有的参数,根据当前的输入计算一个输出
        model_output = model(data)
        # 5.1 用所设计算损失(误差)函数计算误差
        loss = loss_fn(model_output , label)
        if i%500 == 0:
            print(loss)
        # 5.2 每次训练前清零之前计算的梯度(导数)
        optimizer.zero_grad()
        # 5.3 根据误差反向传播计算误差对各个权重的导数
        loss.backward()
        # 5.4 根据优化器里面的算法自动调整神经网络权重
        optimizer.step()

# 保存下训练好的模型,省得下次再重新训练
torch.save(model,'./my_handwrite_recognize_model.pt')
    

##########现在你已经训练好了#################
# 6. 用这个神经网络解决你的问题,比如手写数字识别,输入一个图片矩阵,然后模型返回一个数字
testdataiter = iter(testsetloader)
testimages, testlabels = testdataiter.next()

img_vector = testimages[0].squeeze().view(1,-1)
# 模型返回的是一个1x10的矩阵,第几个元素值最大那就是表示模型认为当前图片是数字几
result_digit = model(img_vector)
print("该手写数字图片识别结果为:", result_digit.max(1)[1],"标签为:",testlabels[0])

评估

import torch
import torchvision
from torchvision import datasets, transforms
# 1. 加载MNIST手写数字数据集数据和标签
transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.5, ), (0.5, ))])
trainset = datasets.MNIST(root='./data', train=True,
                            download=True, transform=transform)
trainsetloader = torch.utils.data.DataLoader(trainset, batch_size=20000, shuffle=True)

testset = datasets.MNIST(root='./data', train=True,
                            download=True, transform=transform)
testsetloader = torch.utils.data.DataLoader(testset, batch_size=20000, shuffle=True)

#######如果你不放心数据有没有加载出可以将图片显示出来看下#######
# dataiter = iter(trainsetloader)
# images, labels = dataiter.next()
# import numpy as np
# import matplotlib.pyplot as plt
# plt.imshow(images[0].numpy().squeeze())
# plt.show()
# print(images.shape)
# print(labels.shape)
##########上面这段是显示图片的代码#############


# 2. 设计网络结构
first_in, first_out, second_out = 28*28,  128, 10
model = torch.nn.Sequential(
    torch.nn.Linear(first_in, first_out),
    torch.nn.ReLU(),
    torch.nn.Linear(first_out, second_out),
)

# 3. 设计损失函数
loss_fn = torch.nn.CrossEntropyLoss()

# 4. 设置用于自动调节神经网络参数的优化器
learning_rate = 1e-4
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)

# 5. 训练神经网络(重复训练10次)
for t in range(10):
    for i, one_batch in enumerate(trainsetloader,0):
        data,label = one_batch
        data[0].view(1,784)# 将28x28的图片变成784的向量
        data = data.view(data.shape[0],-1)

        # 让神经网络根据现有的参数,根据当前的输入计算一个输出
        model_output = model(data)
        # 5.1 用所设计算损失(误差)函数计算误差
        loss = loss_fn(model_output , label)
        if i%500 == 0:
            print(loss)
        # 5.2 每次训练前清零之前计算的梯度(导数)
        optimizer.zero_grad()
        # 5.3 根据误差反向传播计算误差对各个权重的导数
        loss.backward()
        # 5.4 根据优化器里面的算法自动调整神经网络权重
        optimizer.step()

# 保存下训练好的模型,省得下次再重新训练
torch.save(model,'./my_handwrite_recognize_model.pt')
    

##########现在你已经训练好了#################
# 6. 评估神经网络
count = 0
right_count = 0
for i, one_batch in enumerate(testsetloader ,0):
        data,label = one_batch
        count += data.shape[0]
        model_output = model(data)
        right_count += torch.sum(model_output.max(0)[1]==label)
        
print('测试集准确率为:',right_count/count)