pytorch：Logistic回归

一时失言乱红尘 2022-01-31 07:47 495阅读 0赞 
  1. import matplotlib.pyplot as plt
  2. from torch import nn, optim
  3. import numpy as np
  4. import torch
  5. from torch.autograd import Variable
  6. # 设置字体为中文
  7. plt.rcParams['font.sans-serif'] = ['SimHei']
  8. plt.rcParams['axes.unicode_minus'] = False
  9. # 读取数据
  10. def readText():
  11.     with open('data.txt', 'r') as f:
  12.         data_list = f.readlines()
  13.         data_list = [i.split('\n')[0] for i in data_list]
  14.         data_list = [i.split(',') for i in data_list]
  15.         data = [(float(i[0]), float(i[1]), int(i[2])) for i in data_list]
  16.         x_data = [[float(i[0]), float(i[1])] for i in data_list]
  17.         y_data = [float(i[2]) for i in data_list]
  18.     return data, x_data, y_data
  19. # 原始数据可视化
  20. def visualize(data):
  21.     x0 = list(filter(lambda x: x[-1] == 0.0, data))  # 找出类别为0的数据集
  22.     x1 = list(filter(lambda x: x[-1] == 1.0, data))  # 找出类别为1的数据集
  23.     plot_x0_0 = [i[0] for i in x0]  # 类别0的x
  24.     plot_x0_1 = [i[1] for i in x0]  # 类别0的y
  25.     plot_x1_0 = [i[0] for i in x1]  # 类别1的x
  26.     plot_x1_1 = [i[1] for i in x1]  # 类别1的y
  27.     plt.plot(plot_x0_0, plot_x0_1, 'ro', label='类别 0')
  28.     plt.plot(plot_x1_0, plot_x1_1, 'bo', label='类别 1')
  29.     plt.legend()  # 显示图例
  30.     plt.title(label='原始数据分布情况')
  31.     plt.show()
  32. def visualize_after(data):
  33.     x0 = list(filter(lambda x: x[-1] == 0.0, data))  # 找出类别为0的数据集
  34.     x1 = list(filter(lambda x: x[-1] == 1.0, data))  # 找出类别为1的数据集
  35.     plot_x0_0 = [i[0] for i in x0]  # 类别0的x
  36.     plot_x0_1 = [i[1] for i in x0]  # 类别0的y
  37.     plot_x1_0 = [i[0] for i in x1]  # 类别1的x
  38.     plot_x1_1 = [i[1] for i in x1]  # 类别1的y
  39.     plt.plot(plot_x0_0, plot_x0_1, 'ro', label='类别 0')
  40.     plt.plot(plot_x1_0, plot_x1_1, 'bo', label='类别 1')
  41.     # 绘制分类函数
  42.     w0, w1 = model.lr.weight[0]
  43.     w0 = w0.data.item()
  44.     w1 = w1.data.item()
  45.     b = model.lr.bias.item()
  46.     plot_x = np.arange(30, 100, 0.1)
  47.     plot_y = (-w0 * plot_x - b) / w1
  48.     plt.plot(plot_x, plot_y, 'yo', label='分类线')
  49.     plt.legend()  # 显示图例
  50.     plt.title(label='分类线可视化')
  51.     plt.show()
  52. # Logistic回归模型
  53. class LogisticRegression(nn.Module):
  54.     def __init__(self):
  55.         super(LogisticRegression, self).__init__()
  56.         self.lr = nn.Linear(2, 1)
  57.         self.sm = nn.Sigmoid()  # 激活函数类型为sigmoid,经过激活函数，值控制在0到1之间
  58.     def forward(self, x):
  59.         x = self.lr(x)
  60.         x = self.sm(x)
  61.         return x
  62. # 主函数
  63. if __name__ == '__main__':
  64.     data, x_data, y_data = readText()  # 读取数据
  66.     x_data = torch.from_numpy(np.array(x_data))
  67.     y_data = torch.from_numpy(np.array(y_data))
  69.     visualize(data)  # 可视化（观察原始数据分布情况）
  70.     # 获取到模型
  71.     model = LogisticRegression()
  72.     if torch.cuda.is_available():
  73.         model.cuda()
  74.     # 损失函数以及梯度下降
  75.     criterion = nn.BCELoss()  # 二分类的损失函数
  76.     optimizer = optim.SGD(model.parameters(), lr=1e-3, momentum=0.9)
  78.     ctn = []
  79.     lo = []
  80.     for epoch in range(50000):
  81.         if torch.cuda.is_available():
  82.             x = Variable(x_data).cuda()
  83.             y = Variable(y_data).cuda()
  84.         else:
  85.             x = Variable(x_data)
  86.             y = Variable(y_data)
  88.         x = torch.tensor(x, dtype=torch.float32)
  89.         out = model(x)
  90.         y = torch.tensor(y, dtype=torch.float32)
  91.         loss = criterion(out, y)
  92.         print_loss = loss.data.item()
  93.         mask = out.ge(0.5).float()  # 大于0.5则输出为1
  94.         correct = (mask == y).sum()  # 统计输出为1的个数
  95.         acc = correct.item() / x.size(0)
  97.         optimizer.zero_grad()  # 梯度清零
  98.         loss.backward()  # 反向传播
  99.         optimizer.step()  # 更新梯度
  101.         ctn.append(epoch+1)
  102.         lo.append(print_loss)
  104.         if (epoch+1) % 1000 == 0:
  105.             print('*'*10)
  106.             print('epoch {}'.format(epoch+1))
  107.             print('loss is {:.4f}'.format(print_loss))
  108.             print('acc is {:.4f}'.format(acc))
  109.     visualize_after(data)
  110.     # 绘制训练次数与损失值之间的关系
  111.     plt.plot(ctn,lo)
  112.     plt.title(label='训练次数与损失值关系')
  113.     plt.xlabel('训练次数')
  114.     plt.ylabel('损失值')
  115.     plt.show()

实验数据：

  1. 34.62365962451697,78.0246928153624,0
  2. 30.2867107622687,43.89499752400101,0
  3. 35.84740876993872,72.90219802708364,0
  4. 60.18259938620976,86.3855209546826,1
  5. 79.0327360507101,75.3443764369103,1
  6. 45.08327747668339,56.3163717815305,0
  7. 61.10666453684766,96.51142588489624,1
  8. 75.02474556738889,46.55401354116538,1
  9. 76.09878670226257,87.42056971926803,1
  10. 84.43281996120035,43.53339331072109,1
  11. 95.86155507093572,38.22527805795094,0
  12. 75.01365838958247,30.60326323428011,0
  13. 82.30705337399482,76.48196330235604,1
  14. 69.36458875970939,97.71869196188608,1
  15. 39.53833914367223,76.03681085115882,0
  16. 53.9710521485623,89.20735013750265,1
  17. 69.07014406283025,52.74046973016765,1
  18. 67.9468554771161746,67.857410673128,0

实验结果：

watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3Nvbmd4aWFvbGluZ2Jhb2Jhbw_size_16_color_FFFFFF_t_70watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3Nvbmd4aWFvbGluZ2Jhb2Jhbw_size_16_color_FFFFFF_t_70 1watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3Nvbmd4aWFvbGluZ2Jhb2Jhbw_size_16_color_FFFFFF_t_70 2

注意事项：

（1）数据太少，没有找到完整数据。

（2）注意类型转换，float64转为float32，对应代码为：

  1. x = torch.tensor(x, dtype=torch.float32)
  2.  out = model(x)
  3.  y = torch.tensor(y, dtype=torch.float32)

发表评论

表情：
评论列表 （有 0 条评论，495人围观）

还没有评论，来说两句吧...

相关阅读