实现三分类

import torch
import torch.nn.functional as F
import matplotlib.pyplot as plt
import numpy as np

# 生成三分类数据
n_data = torch.ones(100, 2)
x0 = torch.normal(2*n_data, 1) # 类别0的数据，100个样本
y0 = torch.zeros(100) # 类别0的标签为0
x1 = torch.normal(-2*n_data, 1) # 类别1的数据，100个样本
y1 = torch.ones(100) # 类别1的标签为1
x2 = torch.normal(0, 2*n_data) # 类别2的数据，100个样本
y2 = torch.ones(100) * 2 # 类别2的标签为2
x = torch.cat((x0, x1, x2), dim=0).type(torch.FloatTensor) # 将三个类别的数据合并
y = torch.cat((y0, y1, y2), dim=0).type(torch.LongTensor) # 将三个类别的标签合并

# 定义分类模型
class Net(torch.nn.Module):
    def __init__(self, n_feature, n_hidden, n_output):
        super(Net, self).__init__()
        self.hidden = torch.nn.Linear(n_feature, n_hidden)
        self.out = torch.nn.Linear(n_hidden, n_output)

    def forward(self, x):
        x = F.relu(self.hidden(x))
        x = self.out(x)
        return x

net = Net(n_feature=2, n_hidden=40, n_output=3)

# 定义优化器和损失函数
optimizer = torch.optim.SGD(net.parameters(), lr=0.05)
loss_func = torch.nn.CrossEntropyLoss()

# 开始画图
plt.ion()
plt.show()

# 训练模型
for t in range(100):
    out = net(x)
    loss = loss_func(out, y)
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

    # 可视化
    if t % 2 == 0:
        plt.cla()
        pred = torch.max(F.softmax(out), 1)[1] # 得到预测结果
        pred_y = pred.data.numpy().squeeze()
        target_y = y.data.numpy()
        plt.scatter(x.data.numpy()[:, 0], x.data.numpy()[:, 1], c=pred_y, s=100, lw=0, cmap='RdYlBu')
        accuracy = sum(pred_y == target_y) / 300
        plt.text(1, -5, 'Accuracy=%.2f' % accuracy, fontdict={'size': 20, 'color': 'red'})
        plt.pause(0.1)

plt.ioff()
plt.show()