系统学习图像算法Day.37——C++应用——复现机器学习中的“逻辑回归”算法

傷城~ 2023-06-28 11:27 10阅读 0赞

逻辑回归 是为了解决在样本标记为0、1时，希望预测其取1概率 的问题，而不是单纯预测为0或1  
在学习逻辑回归后，自己用C++复现该代码

#include <iostream>
    #include <cmath>
    #include <fstream>
    #include <vector>
    
    using namespace std;
    const int DIMS = 20;
    
    char *file1 = "logical_regression_train.dat";
    char *file2 = "logical_regression_test.dat";
    
    struct training_unit    //每个X数据结构体
    { 
        double x[DIMS+1];
        int y;
    };
    
    vector <training_unit> training;   //用来保存X
    vector <training_unit> test;
    
    int sign( double signx )    //sign函数，正为1，负为-1
    { 
        int signy; 
        if (signx > 0) signy = 1;
        else signy = -1;
        return signy;
    }
    
    double sigmoid( double x )
    { 
        double y;
        y = 1.0/(1+exp(-x));
        return y;
    }
    
    void arr_b2a( double *a, double *b, int Num )
    { 
        for (int i=0; i<Num; i++)
        { 
             a[i] = b[i];
        }
    }
    
    double vector_multiply( const double *a, const double *b, int Num )
    { 
        double ab = 0.0;
        for(int i=0; i<Num; i++)
        { 
           ab += a[i] * b[i];
        }
        return ab;
    }
    
    double errlog ( const vector <training_unit> training, double *W, int N_data )
    { 
        double err = 0.0;
        double temp = 0.0;
        for( int i=0; i<N_data; i++ )
        { 
            temp += log( exp( 1 + exp(-1 * vector_multiply(training[i].x, W, DIMS+1) * (double)training[i].y ) ) );
        }
        err = temp/N_data;
        return err;
    }
    
    double err01(const vector <training_unit> training, double *W, int N_data)
    {    
        double error = 0.0; 
        for(int i = 0; i < N_data; i++)
        {  
            if(sign(vector_multiply(training[i].x,W,DIMS+1)) != training[i].y)  error++;
        } 
        return double(error / N_data);
    }
    
    void grad_Ein( const vector <training_unit> training, double *W, int N_data, double *grad)
    { 
        double temp1 = 0.0;
        for(int i=0; i<N_data; i++)
        { 
            temp1 = sigmoid( -1 * vector_multiply(training[i].x, W, DIMS+1) * (double)training[i].y );
            for(int j=0; j<DIMS+1; j++)
            { 
                grad[j] += temp1*(-1.0 * training[i].y * training[i].x[j]);
            }
         }
        for(int i=0; i<DIMS+1; i++)
        grad[i] = grad[i]/N_data;
    }
    
    void iter_W(double *W, double iter_step, double *grad_Ein_Wt)
    { 
        double W_tadd1[DIMS+1];
        for(int i=0; i<DIMS+1; i++)
        { 
            W_tadd1[i] -= (iter_step * grad_Ein_Wt[i]);
        }
     }
     
    void file_to_vector( char *filename, vector <training_unit>& vectorname )  //将数据文件中的数据读取到vector中去
    { 
        ifstream inFile;
        inFile.open(filename);
        if (!inFile.is_open())
        { 
            cerr <<"Could not open"<<filename<<endl;
            exit(EXIT_FAILURE);
        }
        training_unit x21y1_model;
        while (!inFile.eof())      //把x、y数据读入vector容器
        { 
            x21y1_model.x[0] = 1; 
            for(int i=1;i<DIMS+1;i++)
            { 
                inFile >> x21y1_model.x[i] ;
                inFile >> x21y1_model.y;
            }
         vectorname.push_back(x21y1_model); 
         }
     inFile.close();
    }
    
    void logical_regression( const vector <training_unit> training, const vector <training_unit> test, double iter_step, double *W_init, int iter )
    { 
        int N_data1 = training.size();
        int N_data2 = test.size();
        double W[DIMS+1];
        double err1 = 0.0;
        double err2 = 0.0;
        double grad[DIMS+1] = { 0.0};
        for ( int iter_i=0; iter_i<iter; iter_i++ )
        { 
            if (iter_i == 0) arr_b2a(W, W_init, DIMS+1 );
            grad_Ein( training, W, N_data1, grad );
            iter_W( W, iter_step, grad );
        }
        err1 = errlog( test, W, N_data2 );
        err2 = err01( test, W, N_data2 );
        cout << "The errorlog is " << err1 <<endl;
        cout << "The error01 is " << err2 <<endl;
    }
    
    int main()
    { 
        int iter = 4000;
        double iter_step = 0.001;
        double W_init[DIMS+1];
        for( int i=0; i<DIMS+1; i++ )
            W_init[i] = 1;
        file_to_vector( file1, training );
        file_to_vector( file2, test );
        logical_regression( training, test, iter_step, W_init, iter );
        cin.get();
        return 0;
    }