Pytorch学习记录-逻辑回归

新建 Microsoft PowerPoint 演示文稿 (2).jpg
Pytorch学习记录-逻辑回归

1. 引入必须库&设定超参数

一样的套路

import torch
import torchvision
import torchvision.transforms as transforms
import torch.nn as nn

# 超参数
input_size = 784
num_classes = 10
num_epochs = 5
batch_size = 100
learning_rate = 0.01

2. 获取数据和加载数据

train_dataset = torchvision.datasets.MNIST(root='./data',
                                           train=True,
                                           transform=transforms.ToTensor(),
                                           download=True)
test_dataset = torchvision.datasets.MNIST(root='./data',
                                          train=False,
                                          transform=transforms.ToTensor())

train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
                                           batch_size=batch_size,
                                           shuffle=True)
test_loader = torch.utils.data.DataLoader(dataset=train_dataset,
                                          batch_size=batch_size,
                                          shuffle=False)

3. 构建逻辑回归模型

这里有一个问题，为什么使用Linear之后没有用softmax？
答案就在损失函数，这里的损失函数使用的是CrossEntropyLoss()，多分类用的交叉熵损失函数，用这个 loss 前面不需要加 Softmax 层。
我重新写了一个Model类，但是使用MSELoss等损失函数都会报错

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.linear = nn.Linear(input_size, num_classes)
        self.sigmoid = nn.Sigmoid()

    def forward(self, x):
        y_pred = self.sigmoid(self.linear(x))
        return y_pred


model = Model()
criterion = nn.MSELoss()
# model = nn.Linear(input_size, num_classes)
# criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate)

4. 训练模型

total_step = len(train_loader)
for epoch in range(num_epochs):
    for i, (images, labels) in enumerate(train_loader):
        images = images.reshape(-1, 28 * 28)
        outputs = model(images)
        loss = criterion(outputs, labels)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if (i + 1) % 100 == 0:
            print('Epoch [{}/{}], Step [{}/{}] ,Loss:{:.5f}'.format(epoch + 1, num_epochs, i + 1, total_step,
                                                                    loss.item()))

5. 测试模型并保存模型

with torch.no_grad():
    correct = 0
    total = 0
    for images, labels in test_loader:
        images = images.reshape(-1, 28 * 28)
        outputs = model(images)
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum()
    print('Accuracy of the model on the 10000 test images: {} %'.format(100 * correct / total))

torch.save(model.state_dict(),'LogisticModel.ckpt')

Epoch [1/5], Step [100/600] ,Loss:1.58177
Epoch [1/5], Step [200/600] ,Loss:1.22867
Epoch [1/5], Step [300/600] ,Loss:1.18859
Epoch [1/5], Step [400/600] ,Loss:0.79991
Epoch [1/5], Step [500/600] ,Loss:0.93516
Epoch [1/5], Step [600/600] ,Loss:0.74280
Epoch [2/5], Step [100/600] ,Loss:0.66908
Epoch [2/5], Step [200/600] ,Loss:0.78997
Epoch [2/5], Step [300/600] ,Loss:0.69782
Epoch [2/5], Step [400/600] ,Loss:0.57147
Epoch [2/5], Step [500/600] ,Loss:0.62612
Epoch [2/5], Step [600/600] ,Loss:0.52377
Epoch [3/5], Step [100/600] ,Loss:0.52046
Epoch [3/5], Step [200/600] ,Loss:0.65288
Epoch [3/5], Step [300/600] ,Loss:0.45212
Epoch [3/5], Step [400/600] ,Loss:0.57578
Epoch [3/5], Step [500/600] ,Loss:0.46629
Epoch [3/5], Step [600/600] ,Loss:0.51632
Epoch [4/5], Step [100/600] ,Loss:0.52777
Epoch [4/5], Step [200/600] ,Loss:0.51157
Epoch [4/5], Step [300/600] ,Loss:0.50300
Epoch [4/5], Step [400/600] ,Loss:0.48730
Epoch [4/5], Step [500/600] ,Loss:0.49723
Epoch [4/5], Step [600/600] ,Loss:0.36077
Epoch [5/5], Step [100/600] ,Loss:0.43910
Epoch [5/5], Step [200/600] ,Loss:0.58018
Epoch [5/5], Step [300/600] ,Loss:0.56438
Epoch [5/5], Step [400/600] ,Loss:0.33580
Epoch [5/5], Step [500/600] ,Loss:0.39300
Epoch [5/5], Step [600/600] ,Loss:0.54957
Accuracy of the model on the 10000 test images: 88 %