随机梯度下降算法的值误差

问题描述

我无法运行我的 SGD 代码，我不知道问题出在哪里。如果你能帮助我，那就太好了。这是我的代码：

class logistic_regression: 
    
    def __init__(self,X,y_actual,alpha,max_iter,batch_size):
        self.X = X 
        self.y_actual = y_actual 
        self.alpha = alpha 
        self.max_iter = max_iter   
        self.batch_size = batch_size
    
    def sigmoid(self,z): 
        return 1/(1+np.exp(-z))
    
    def predictor(self,theta,X): 
        predictions = np.matmul(X,theta) 
        sigmoidal_prediction = self.sigmoid(predictions) 
        return(sigmoidal_prediction)
    
    def loss(self,h,y):
        h = h + 1e-9
        h = np.array(h,dtype=np.complex128) 
        y = np.array(y,dtype=np.complex128)
        h = h.flatten()
        y = y.flatten()
        return (-((y*np.log(h))-((1-y)*np.log(1-h)))).mean()
    
    def stochastic_gradient_descent(self): 
        X1 = np.matrix(sm.add_constant(self.X))  
        m,n = X1.shape   
        y_actual = self.y_actual.to_numpy().reshape(m,1)  
        Xy = np.c_[X1,y_actual] 
        
        # Initializing the random number generator   
        rng = np.random.default_rng(seed=123)    
        
        theta = np.ones((n,1)) 
        predictions = None
        for i in range(0,self.max_iter): 
            rng.shuffle(Xy) # Shuffle X and y
            
            # Performing minibatch moves
            for i in range(self.batch_size):
                j = i + self.batch_size
                X_batch,y_batch = Xy[i:j,:-1],Xy[i:j,-1:] 
                predictions = self.predictor(theta,X_batch) 
                gradient = np.matmul(np.transpose(X_batch),(predictions-y_batch))/self.batch_size
                
                theta = theta - self.alpha*gradient
        
        f1 = metrics.f1_score(y_actual,np.around(predictions),labels=None,pos_label=1,average='binary',sample_weight=None)
        ceo = self.loss(predictions,y_actual)
        print("\nCross Entropy: %f" % (ceo),"\nAlpha = %s" % self.alpha,"\nIterations: %s" % self.max_iter,"\nF1 Score: ",f1)
        return(theta) 
              
    def classifier(self,threshold=0.5):
        X1 = np.matrix(sm.add_constant(self.X))
        theta = self.stochastic_gradient_descent()
        return [1 if i >= threshold else 0 for i in self.predictor(theta,X1)]

我调用这个函数：

log_reg = logistic_regression(X_test_std,y_test,0.01,100,2)
print(log_reg.classifier())

但是出现了值错误：

ValueError: 发现输入变量的样本数量不一致：[1151,2]

尺寸问题位于 f1 中的 ceo 和 def stochastic_gradient_descent(self)。但我不知道如何解决这个问题。你能给我一些提示吗？

解决方法

暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！

如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。

小编邮箱:dio#foxmail.com (将#修改为@）