Pytorch基础-logistic回归

import torch
import torch.nn as nn
import numpy as np

print(torch.__version__)

1.1.0

Logistic回归

logistic回归是一中广义线性回归，与多重线性回归分析有很多相同之处。它们模型形式上基本相同，都具有wx+b,其中w和b是待求解参数，区别在于因变量不同，多重线性回归直接将wx+b作为因变量，即y=wx+b，而logistic回归则通过函数L将wx+b对应一个隐状态p，p=L(wx+b)，根据p与1-p的大小决定因变量的值。

L为logistic函数时为logistic回归，L为多项式函数时为多项式回归

logistic回归主要进行二分类预测：sigmoid函数就是常见的logistic函数，因为sigmoid函数的输出时0~1之间的概率，当概率大于0.5时预测为1，小于0.5时为0

# 加载数据
data = np.loadtxt('./data/german.data-numeric')
n, l = data.shape
# 数据归一化
for j in range(l - 1):
    meanVal = np.mean(data[:, j])
    stdVal = np.std(data[:, j])
    data[:, j] = (data[:, j] - meanVal) / stdVal

# 打乱数据
np.random.shuffle(data)

# 划分训练集和测试集
train_data = data[:900, :l - 1]
train_tag = data[:900, l - 1] - 1
test_data = data[900:, :l - 1]
test_tag = data[900:, l - 1] - 1

# 定义网络
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.fc = nn.Linear(24, 2)
        self.sigmoid = nn.Sigmoid()

    def forward(self, x):
        out = self.fc(x)
        out = self.sigmoid(out)
        return out

def test(pred, lab):
    t = pred.max(-1)[1] == lab
    return torch.mean(t.float())

net = Net()
critertion = nn.CrossEntropyLoss()  # 使用交叉熵损失
optimizer = torch.optim.Adam(net.parameters())  # Adam优化器
epochs = 1000

for epoch in range(epochs):
    net.train()  # 指定模型为训练模型，计算梯度
    x = torch.from_numpy(train_data).float()
    y = torch.from_numpy(train_tag).long()
    y_pred = net(x)
    loss = critertion(y_pred, y)  # 计算损失
    optimizer.zero_grad()  # 权重置零
    loss.backward()  # 反向传播
    optimizer.step()

    if (epoch + 1) % 100 == 0:
        net.eval()  #指定模型计算模式
        test_in = torch.from_numpy(test_data).float()
        test_t = torch.from_numpy(test_tag).long()
        test_out = net(test_in)
        # 使用测试函数计算准确率
        accu = test(test_out, test_t)
        print('epoch: {}, loss: {}, accuracy:{}'.format(
            epoch + 1, loss.item(), accu))

epoch: 100, loss: 0.6658604145050049, accuracy:0.699999988079071
epoch: 200, loss: 0.6306068897247314, accuracy:0.8199999928474426
epoch: 300, loss: 0.6095178723335266, accuracy:0.8100000023841858
epoch: 400, loss: 0.5955496430397034, accuracy:0.8100000023841858
epoch: 500, loss: 0.5853410363197327, accuracy:0.800000011920929
epoch: 600, loss: 0.5774123072624207, accuracy:0.8199999928474426
epoch: 700, loss: 0.5710282921791077, accuracy:0.8199999928474426
epoch: 800, loss: 0.5657661557197571, accuracy:0.8199999928474426
epoch: 900, loss: 0.5613517165184021, accuracy:0.8199999928474426
epoch: 1000, loss: 0.5575944185256958, accuracy:0.8199999928474426