推荐系统偏好SVD C++

最近想整整推荐系统，比较经典的算法就是SVD了。具体理论不多讲了。直接上代码。

先贴张效果图吧。userNum 6040 itemNum 3900

本文链接：http://www.cnblogs.com/wn19910213/p/3617781.html

上代码咯：

SVD.h

#ifndef SVD_H_INCLUDED
#define SVD_H_INCLUDED

#include <vector>
#include <cstring>

using namespace std;

class SVD{
    public:
        SVD(double*,double*,int,double**,double**);
        ~SVD();

        void loadTrainFile(string);
        double predictScore(int,double,double,double*,double*);
        double Validate(string,double,double*,double*,double**,double**);
//    private:
        double* Bi;
        double* Bu;
        int factor;
        double** Qi;
        double** Pu;
};


#endif // SVD_H_INCLUDED

SVD.cpp

#include <cmath>
#include <iostream>
#include <cstring>
#include <cstdlib>
#include <fstream>
#include "SVD.h"


int userNum = 6040;
int itemNum = 3900;
double AVG = 3.579231;
double lr = 0.01;
double theta = 0.05;
double preRmse = 1000000.0;

int main()
{
    string trainFile = "/home/ja/CADATA/SVD/ml_data/training.txt";
    string testFile = "/home/ja/CADATA/SVD/ml_data/test.txt";
    srand(0);
    SVD svd(NULL,NULL,0,NULL,NULL);

    for(size_t i=0;i<100;i++){
        svd.loadTrainFile(trainFile);
        //lr *= 0.9;
        double curRmse = svd.Validate(testFile,AVG,svd.Bu,svd.Bi,svd.Pu,svd.Qi);
        cout << "test_Rmse in step " << i << ": " << curRmse << endl;
        if(curRmse >= preRmse){
            break;
        }
        else{
            preRmse = curRmse;
        }
    }
    return 0;
}

double SVD::Validate(string testfile,double avg,double* bu,double* bi,double** pu,double** qi){
    ifstream fin(testfile.c_str());
    if(!fin){
        cout << "error" << endl;
    }
    int userId,itemId,rating,t;
    int n = 0;
    double rmse;
    while(fin >> userId >> itemId >> rating >> t){
        n++;
        double pScore = predictScore(avg,bu[userId-1],bi[itemId-1],pu[userId-1],qi[itemId-1]);
        rmse += (rating - pScore) * (rating - pScore);
    }
    fin.close();
    return sqrt(rmse/n);
}

double SVD::predictScore(int avg,double bu,double bi,double* pu,double* qi){
    double tmp = 0.0;
    for(size_t i=0;i<factor;i++){
        tmp += pu[i] * qi[i];
    }

    double score = avg + bu + bi + tmp;
    if(score > 5){
        score = 5;
    }
    if(score < 1){
        score = 1;
    }
    return score;
}

void SVD::loadTrainFile(string file){
    ifstream fin(file.c_str());
    if(!fin){
        cout << "error" << endl;
    }

    int userId,itemId,rating,t;
    while(fin >> userId >> itemId >> rating >> t){
        double predict = predictScore(AVG,Bu[userId-1],Bi[itemId-1],Pu[userId-1],Qi[itemId-1]);
        double error = rating - predict;
        Bu[userId-1] += lr * (error - theta * Bu[userId-1]);
        Bi[itemId-1] += lr * (error - theta * Bi[itemId-1]);

        for(size_t i=0;i<factor;i++){
            double Tmp = Pu[userId-1][i];
            Pu[userId-1][i] += lr * (error * Qi[itemId-1][i] - theta * Pu[userId-1][i]);
            Qi[itemId-1][i] += lr * (error * Tmp - theta * Qi[itemId-1][i]);
        }
    }
    fin.close();
}

SVD::SVD(double* bi,double* bu,int k,double** qi,double** pu){

    if(bi == NULL){
        Bi = new double[itemNum];
        for(size_t i=0;i<itemNum;i++){
            Bi[i] = 0.0;
        }
    }
    else{
        Bi = bi;
    }

    if(bu == NULL){
        Bu = new double[userNum];
        for(size_t i=0;i<userNum;i++){
            Bu[i] = 0.0;
        }
    }
    else{
        Bu = bu;
    }

    factor = 10;

    if(qi == NULL){
        Qi = new double* [itemNum];
        for(size_t i=0;i<itemNum;i++){
            Qi[i] = new double[factor];
        }

        for(size_t i=0;i<itemNum;i++){
            for(size_t j=0;j<factor;j++){
                Qi[i][j] = 0.1 * (rand() / (RAND_MAX + 1.0)) / sqrt(factor);
            }
        }
    }
    else{
        Qi = qi;
    }

    if(pu == NULL){
        Pu = new double* [userNum];
        for(size_t i=0;i<userNum;i++){
            Pu[i] = new double[factor];
        }

        for(size_t i=0;i<userNum;i++){
            for(size_t j=0;j<factor;j++){
                Pu[i][j] = 0.1 * (rand() / (RAND_MAX + 1.0)) / sqrt(factor);
            }
        }
    }
    else{
        Pu = pu;
    }
}

SVD::~SVD(){
    delete[] Bi;
    delete[] Bu;
    for(size_t i=0;i<userNum;i++){
        delete[] Pu[i];
    }
    for(size_t i=0;i<itemNum;i++){
        delete[] Qi[i];
    }
    delete[] Pu;
    delete[] Qi;
}