#include <iostream>
#include <string>
#include <fstream>
#include <sstream>
#include <vector>
#include <cmath>
template <typename DataType>
double sigmoid(DataType z) {
return 1.0/(1+exp((-1)*z));
}
template <typename DataType, typename WeightType>
double getMatResult(typename::std::vector<DataType> &data, typename::std::vector<WeightType> &weights) {
double result=0.0;
for(size_t i=0;i<data.size();++i) {
result+=data.at(i)*weights.at(i);
}
return result;
}
template <typename DataType>
void DisplayData(typename::std::vector<std::vector<DataType> > &vv) {
std::cout<<"the number of data: "<<vv.size()<<std::endl;
for(size_t i=0;i<vv.size();++i) {
for(typename::std::vector<DataType>::iterator it=vv[i].begin();it!=vv[i].end();++it) {
std::cout<<*it<<" ";
}
std::cout<<std::endl;
}
}
template <typename DataType, typename WeightType>
double CostFun(typename::std::vector<std::vector<DataType> > &vv, typename::std::vector<WeightType> &v_weights) {
double J=0.0;
typename::std::vector<DataType> v_x;
for(size_t i=0;i<vv.size();++i) {
v_x.clear();
v_x.push_back(vv[i][0]);
v_x.push_back(vv[i][1]);
v_x.push_back(vv[i][2]);
double z=getMatResult(v_x,v_weights);
J=J+vv[i][3]*log2(sigmoid(z))+(1-vv[i][3])*log2(1-sigmoid(z));
}
J=-J/vv.size();
return J;
}
int main() {
std::ifstream infile_feat("train.data");
std::string feature;
float feat_onePoint;
std::vector<float> lines;
std::vector<double> v_weights;
std::vector<std::vector<float> > lines_feat;
lines_feat.clear();
v_weights.clear();
for(size_t i=0;i<3;++i) {
v_weights.push_back(1.0);
}
while(!infile_feat.eof()) {
getline(infile_feat, feature);
if(feature.empty())
break;
std::stringstream stringin(feature);
lines.clear();
lines.push_back(1.0);
while(stringin >> feat_onePoint) {
lines.push_back(feat_onePoint);
}
lines_feat.push_back(lines);
}
infile_feat.close();
std::cout<<"display train data: "<<std::endl;
DisplayData(lines_feat);
double res=CostFun(lines_feat, v_weights);
std::cout<<"the value of cost function: "<<res<<std::endl;
std::vector<double> v_x;
while(true) {
double grad0=0.0,grad1=0.0,grad2=0.0;
for(size_t i=0;i<lines_feat.size();++i) {
v_x.clear();
v_x.push_back(lines_feat[i][0]);
v_x.push_back(lines_feat[i][1]);
v_x.push_back(lines_feat[i][2]);
grad0+=(lines_feat[i][3]-sigmoid(getMatResult(v_x,v_weights)))*lines_feat[i][0];
grad1+=(lines_feat[i][3]-sigmoid(getMatResult(v_x,v_weights)))*lines_feat[i][1];
grad2+=(lines_feat[i][3]-sigmoid(getMatResult(v_x,v_weights)))*lines_feat[i][2];
}
grad0=grad0/lines_feat.size();
grad1=grad1/lines_feat.size();
grad2=grad2/lines_feat.size();
//0.03为学习率阿尔法
v_weights[0]=v_weights[0]+0.03*grad0;
v_weights[1]=v_weights[1]+0.03*grad1;
v_weights[2]=v_weights[2]+0.03*grad2;
double res_new;
res_new=CostFun(lines_feat,v_weights);
if(std::abs(res_new-res)<0.0000000001)
break;
res=res_new;
}
for(size_t i=0;i<3;++i) {
std::cout<<v_weights.at(i)<<" ";
}
std::cout<<std::endl;
lines_feat.clear();
infile_feat.open("test.data");
while(!infile_feat.eof()) {
getline(infile_feat, feature);
if(feature.empty())
break;
std::stringstream stringin(feature);
lines.clear();
lines.push_back(1.0);
while(stringin >> feat_onePoint) {
lines.push_back(feat_onePoint);
}
lines_feat.push_back(lines);
}
infile_feat.close();
std::cout<<"display test data: "<<std::endl;
DisplayData(lines_feat);
for(size_t i=0;i<lines_feat.size();++i) {
v_x.clear();
v_x.push_back(lines_feat[i][0]);
v_x.push_back(lines_feat[i][1]);
v_x.push_back(lines_feat[i][2]);
res=getMatResult(v_x,v_weights);
double lable=sigmoid(res);
for(size_t j=0;j<4;++j) {
std::cout<<lines_feat[i][j]<<" ";
}
if(lable>0.5)
std::cout<<" 1"<<std::endl;
else
std::cout<<" 0"<<std::endl;
}
return 0;
}
来源：https://www.cnblogs.com/donggongdechen/p/7634628.html