2022.3.1科研记录

2022.3.1科研记录

《pytorch深度学习入门》曾芃壹
3.4逻辑回归
3.最大概似法与交叉熵
抛形状不规则的硬币
似然度：每个样本结果概率的乘积
让似然度P(y|X,w)最大就是让损失函数L(w)最小(因为二者是-log的关系）
↑像这样利用最大概似法得到的损失函数称为交叉熵
交叉熵损失函数是分类问题中常用的损失函数，pytorch预置了nn.CrossEntropyLoss()

4.逻辑回归示例

import torch
import matplotlib.pyplot as plt
from torch import nn, optim

cluster = torch.ones(500, 2)
data0 = torch.normal(4*cluster, 2)
data1 = torch.normal(-4*cluster, 2)
label0 = torch.zeros(500)  # 生成500个0标签
label1 = torch.ones(500)  # 生成500个1标签

x = torch.cat((data0, data1),).type(torch.FloatTensor)  # 将data0,data1合并到一起
y = torch.cat((label0, label1),).type(torch.LongTensor)  # 将label0,label1合并到一起

plt.scatter(x.numpy()[:, 0], x.numpy()[:, 1], c=y.numpy(), s=10, lw=0, cmap='RdYlGn')  # 0标签的是红色，1标签的是绿色
plt.show()


class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.linear = nn.Linear(2, 2)

    def forward(self, x):
        # 激活函数处理数据
        x = self.linear(x)
        x = torch.sigmoid(x)
        return x


CUDA = torch.cuda.is_available()

if CUDA:
    net = Net().cuda()
    inputs = x.cuda()
    target = y.cuda()
else:
    net = Net()
    inputs = x
    target = y

optimizer = optim.SGD(net.parameters(), lr=0.02)
criterion = nn.CrossEntropyLoss()


def draw(output):
    if CUDA:
        output = output.cpu()  # 若使用了CUDA加速这一步要还原为CPU数据类型
    plt.cla()
    output = torch.max((output),1)[1]
    pred_y = output.data.numpy().squeeze()
    target_y = y.numpy()
    plt.scatter(x.numpy()[:,0],x.numpy()[:,1],c=y.numpy(),s=10,lw=0,cmap='RdYlGn')
    accuracy = sum(pred_y == target_y)/1000.0
    plt.text(1.5, -4, 'Accuracy=%s' % (accuracy), fontdict={'size': 20, 'color': 'red'})
    plt.pause(0.1)


def train(model, criterion, optimizer, epochs):  # 训练
    for epoch in range(epochs):
        output = model(inputs)  # inputs输入神经网络模型
        loss = criterion(output, target)  # 计算损失函数
        optimizer.zero_grad()  # 清空权重的grad值
        loss.backward()  # 计算梯度
        optimizer.step()  # 进行权值更新
        if epoch % 40 == 0:
            draw(output)


train(net,criterion,optimizer,1000)