使用JAMA lib进行Logistic回归的成本函数

Question

我目前正在使用JAMA lib编写逻辑回归的成本函数。但它不起作用。我不知道为什么。它应该返回一个值：0.6743

public matrix cost() { 
    double[][] sigmoid = sigmoidFunction().getArray(); 
    double[][] sigmoid2 = sigmoidFunction().getArray(); 
    int m = sigmoidFunction().getRowdimension(); 
    int n = sigmoidFunction().getColdimension(); 

    for (int i = 0; i<m; i++) { 
     for (int j =0; j< n; j++) { 
      sigmoid[i][j] = Math.log(sigmoid[i][j]); 
     } 
    } 
    for (int i = 0; i<m; i++) { 
     for (int j =0; j< n; j++) { 
      sigmoid2[i][j] = Math.log(1-sigmoid2[i][j]); 
     } 
    } 

    matrix regularized = theta.transpose().times(theta); 
    double[][] reg = regularized.getArray(); 
    for(int i = 0; i< regularized.getRowdimension(); i++) { 
     for (int j = 0; j< regularized.getColdimension(); j++) { 
      reg[i][j] = lambda/(2*m) * (reg[i][j]); 
     } 
    } 
    regularized = new matrix(reg); 

    matrix log_hx = new matrix(sigmoid); 
    matrix log1_hx = new matrix(sigmoid2); 

    matrix y_1 = Y; 
    y_1 = y_1.transpose().subtract(1); 
    Y = Y.uminus(); 
    Y= Y.transpose(); 
    //J = 1/m * (-y' * log(hx) - (1-y)' * log(1-hx)) 
    matrix J = Y.times(log_hx).subtract(y_1.times(log1_hx)); 

    double [][] cost = J.getArray(); 
    for(int i = 0; i< J.getRowdimension(); i++) { 
     for (int j =0; j< J.getColdimension(); j++) { 
      cost[i][j]= 1/m * cost[i][j]; 
     } 
    } 
    //J = new matrix(cost); 
    //J.addEquals(regularized); 
    return J; 
} 
} 

当我返回如上所示的矩阵J时，它返回0.0。但是当我直接返回Y.times（log_hx）.subtract（y_1.times（log1_hx））时，它神奇地返回值3.3715。当它不是由1/M相乘，并通过正则化

Answer 1

加入这是正确的I弄明白：我删除FF代码：

double [][] cost = J.getArray(); 
for(int i = 0; i< J.getRowdimension(); i++) { 
    for (int j =0; j< J.getColdimension(); j++) { 
     cost[i][j]= 1/m * cost[i][j]; 
    } 

和与此

double[][] cost = J.getArray(); 
double cost_temp = cost[0][0]*1/m; 
J.set_element(0,0,cost_temp); 
J.addEquals(regularized); 
return J; 

取代它现在，我不知道它为什么起作用。