代码:

  1. from sklearn.datasets import load_iris, load_boston
  2. from sklearn.model_selection import train_test_split
  3. import torch
  4. from torch.nn import functional as F
  5. from torch import nn
  6. from sklearn.metrics import accuracy_score, confusion_matrix
  7. import  numpy as np
  10. class NeuralNetworksModel(nn.Module):
  11.     """神经网络模型"""
  13.     def __init__(self, x_dim, y_dim, hidden_size):
  14.         super().__init__()
  16.         self.hidden_1 = nn.Linear(x_dim, hidden_size)
  17.         self.liner = nn.Linear(hidden_size, y_dim)
  19.     def forward(self, x):
  20.         out_1 = F.relu(self.hidden_1(x))
  21.         out = self.liner(out_1)
  22.         return out
  25. class IrisDataDeal():
  26.     """数据处理"""
  27.     def __init__(self):
  28.         self.data = None
  29.         self.target = None
  30.         self.target_names = None
  32.     def get_data(self):
  33.         iris_data = load_iris()
  34.         self.data = iris_data.get('data')
  35.         self.target = iris_data.get('target')
  36.         self.target_names = iris_data.get('target_names')
  37.         return train_test_split(self.data, self.target, test_size=0.2)
  41. class IrisClassify():
  43.     def __init__(self, model, data):
  44.         self.model = model
  45.         self.X_train, self.X_test, self.y_train, self.y_test = data.get_data()
  46.         # 迭代次数
  47.         self.steps = 1000
  48.         # 学习率
  49.         self.learning_rate = 0.01
  52.     def data_trans_form(self):
  53.         X_train = torch.tensor(self.X_train, dtype=torch.float32)
  54.         X_test = torch.tensor(self.X_test, dtype=torch.float32)
  55.         y_train = torch.tensor(self.y_train, dtype=torch.float32)
  56.         y_test = torch.tensor(self.y_test, dtype=torch.float32)
  57.         return X_train, X_test, y_train, y_test
  59.     def train_model(self):
  60.         # 策略: 交叉熵
  61.         loss_func = F.cross_entropy
  62.         op = torch.optim.Adam(self.model.parameters(), lr=self.learning_rate)
  63.         X_train, X_test, y_train, y_test = self.data_trans_form()
  64.         for step in range(self.steps):
  65.             # 前向传播,计算损失
  66.             # 交叉熵损失不需要自己进行one-hot编码
  67.             loss = loss_func(self.model(X_train), y_train.long())
  68.             # 反向传播
  69.             loss.backward()
  70.             # 更新参数
  71.             op.step()
  72.             # 梯度清0
  73.             op.zero_grad()
  74.             if step % 100 == 0:
  75.                 print('loss: %s' % loss.item())
  76.         # 保存模型省略
  77.         # 打印训练集 准确度和混淆矩阵
  78.         outputs = model(X_train)
  79.         predict =  torch.max(outputs.data, 1)[1].cpu()
  80.         train_acc = accuracy_score(y_train.long(), predict)
  81.         print(train_acc)
  82.         confusion = confusion_matrix(y_train.long(), predict)
  83.         print(confusion)
  84.         # 打印测试集 准确度和预测结果与真实结果
  85.         test_outputs = model(X_test)
  86.         predict_test = torch.max(test_outputs.data, 1)[1].cpu()
  87.         test_acc = accuracy_score(y_test.long(), predict_test)
  88.         print(test_acc)
  89.         print(predict_test.numpy())
  90.         print(np.array(y_test.numpy(), dtype=np.int))
  93. if __name__ == '__main__':
  94.     iris_data = IrisDataDeal()
  95.     model = NeuralNetworksModel(4, 3, 4)
  96.     IrisClassify(model, iris_data).train_model()