自定义逻辑回归模型

import torch
import numpy as np
import joblib
from sklearn.model_selection import train_test_split
from matplotlib import pyplot as plt
#⾃定义逻辑回归模型
class LogisticRegression(object):
 
 #定义初始化⽅法
 def __init__(self, max_iters=1000, learning_rate=1e-9):
 
   #参数初始化
   self.max_iters = max_iters
   self.learning_rate = learning_rate
   self.w = None
   self.b = None
   self.n_features = None
   self.losses = []
   self.train_accs = []
   self.test_accs = []
   #保存模型
   self.best_w = None
   self.best_b = None
   self.best_train_acc = -float("inf")
   self.best_test_acc = -float("inf")
 
 #定义逻辑回归⽅法
 def _logic_regression(self, X):
 '''
 逻辑回归的正向传播过程
 '''
   return torch.sigmoid(X @ self.w + self.b)
 
 #定义损失⽅法
 def _get_loss(self, y_true, y_pred):
 '''
 ⽤MSE来衡量误差
 '''
   return ((y_true - y_pred) ** 2).mean()
 
 #定义准确率⽅法
 def _get_acc(self, X, y):
 
   #预测结果
   y_pred = self.predict(X=X)
   y_true = y.view(-1)
   acc = (y_true == y_pred).to(dtype=torch.float32).mean()
   return acc
 
 #定义训练⽅法
 def fit(self, X, y):
 
   self.n_features = X.size(1)
   #初始化w和b
   self.w = torch.randn(size=(self.n_features, 1), requires_grad=True)
   self.b = torch.zeros(size=(1,) ,requires_grad=True)
   #开始迭代，梯度下降
   for step in range(self.max_iters):
     #第⼀步：正向传播
     y_pred = self._logic_regression(X=X)
     #第⼆步：求损失函数
     loss = self._get_loss(y_true=y, y_pred=y_pred)
     self.losses.append(loss.data.cpu().item())
     self.train_accs.append(self._get_acc(X=X_train, y=y_train))
     self.test_accs.append(self._get_acc(X=X_test, y=y_test))
     #第三步：反向传播
     loss.backward()
     #第四步：优化⼀步
     self.w.data -= self.learning_rate * self.w.grad
     self.b.data -= self.learning_rate * self.b.grad
     #清空梯度
     self.w.grad.zero_()
     self.b.grad.zero_()
 
     self.best_train_acc = self.train_accs[-1] if self.train_accs[-1] >       self.best_train_acc else self.best_train_acc
     self.best_test_acc = self.test_accs[-1] if self.test_accs[-1] >    self.best_test_acc else self.best_test_acc
     #第六步：判断是否继续迭代
     if len(self.losses) >= 2:
        if np.abs(self.losses[-1] - self.losses[-2]) <= 1e-6:
           #保存模型
           self.best_w = self.w
           self.best_b = self.b
           print(f"提前终⽌迭代了，总共迭代次数为: {step + 1}")
           break
 
 
#定义推理⽅法
def predict(self, X):
 
    y_pred = self._logic_regression(X=X)
    return (y_pred >= 0.5).to(dtype=torch.long).view(-1)




#导⼊数据
X = []
y = []
with open(file="breast_cancer.csv", mode='r', encoding='utf8') as f:
 f.readline()
 for line in f:
 line = line.strip()
 if line:
 line = [float(ele) for ele in line.split(",")]
 X.append(line[:-1])
 y.append(line[-1])


X = torch.tensor(data=X, dtype=torch.float32)
y = torch.tensor(data=y, dtype=torch.float32).view(-1, 1)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)
#数据预处理，规范化
_mean = X_train.mean(dim=0)
_std = X_train.std(dim=0) + 1e-9
X_train = (X_train - _mean) / _std
X_test = (X_test - _mean) / _std
#构建模型
lr = LogisticRegression(max_iters=100000, learning_rate=1e-2)
#训练模型
lr.fit(X=X_train, y=y_train)
#绘图
plt.plot(lr.losses)
plt.plot(lr.train_accs, label='train_acc')
plt.plot(lr.test_accs, label='test_acc')
plt.legend()
plt.title(label="Accs on Train and Test dataset")
plt.xlabel(xlabel="iters")
plt.ylabel(ylabel="Acc")
#输出最佳模型数据
print(f'最佳的权重w:{lr.best_w},最佳的偏置b:{lr.best_b},最佳的训练集准确率:{lr.best_train_acc},最佳的测试集准确率:{lr.best_test_acc}')

#保存模型
model = joblib.dump(lr, 'logistic_regression.model')

'''
 下⾯模拟在另⼀台⽣产机器上加载模型，并做后续操作
'''
#加载模型
model = joblib.load('logistic_regression.model')
#使⽤best模型进⾏推理
best_model = LogisticRegression(max_iters=100)
best_model.w = model.best_w
best_model.b = model.best_b
y_pred = best_model.predict(X=X_test)
acc = (y_pred == y_test.view(-1)).to(dtype=torch.float32).mean().data.cpu().item()
acc