C++基于特征向量的KNN分类算法

#include "stdafx.h"
#include <iostream>
#include <fstream>
#include <sstream>
#include <string>
#include <vector>
#include <set>
#include <map>
#include <cmath>
 
using namespace std;
 
//样本特征结构体
struct sample
{
 string type;
 vector<double> features;
};
 
//读取训练样本train.txt，训练样本格式：类型名+特征向量
void readTrain(vector<sample>& train, const string& file)
{
 ifstream fin(file.c_str()); //file是存储希望读写的文件名的string对象，fin是读的流
 if(!fin)
 {
 cerr<<"Unable to open the input file: "<<file<<endl;
 exit(1);
 }
 
 string line; 
 double d=0.0;
 while(getline(fin,line)) //fin是读入流，getline从输入流fin读入一行到line
 {
 istringstream stream(line); //bind to stream to the line we read
 sample ts;
 stream>>ts.type;
 while(stream>>d) //read a word from line
 {
  ts.features.push_back(d); //在trains.features的末尾添加一个值为d的元素
 }
 train.push_back(ts); //在train的末尾添加一个值为ts的元素
 }
 fin.close();
}
 
//读取测试样本test.txt，每行都是一个特征向量
void readTest(vector<sample>& test, const string& file)
{
 ifstream fin(file.c_str());
 if(!fin)
 {
 cerr<<"Unable to open the input file: "<<file<<endl;
 exit(1);
 }
 
 string line;
 double d=0.0;
 while(getline(fin,line))
 {
 istringstream stream(line); //bind to stream to the line we read
 sample ts;
 while(stream>>d)
 {
  ts.features.push_back(d);
 }
 test.push_back(ts);
 }
 fin.close();
}
 
//输出结果，为每一个向量赋予一个类型，写入result.txt中
void writeResult(const vector<sample>& test, const string& file)
{
 ofstream fout(file.c_str());
 if(!fout)
 {
 cerr<<"Unable to write the input file: "<<endl;
 exit(1);
 }
 
 for(vector<sample>::size_type i=0;i!=test.size();++i)
 {
 fout << test[i].type << '\t';
 for(vector<double>::size_type j=0;j!=test[j].features.size();++j)
 {
  fout<<test[i].features[j]<<' ';
 }
 fout<<endl;
 }
}
 
//KNN算法的实现
void knnProcess(vector<sample>& test, const vector<sample>& train, const vector<vector<double> >& dm, unsigned int k)
{
 for (vector<sample>::size_type i = 0; i != test.size(); ++i)
 {
 multimap<double, string> dts; //保存与测试样本i距离最近的k个点
 
 for (vector<double>::size_type j = 0; j != dm[i].size(); ++j)
 {
  if (dts.size() < k) //把前面k个插入dts中
  {
  dts.insert(make_pair(dm[i][j], train[j].type)); //插入时会自动排序，按dts中的double排序，最小的排在最后
  }
  else
  {
  multimap<double, string>::iterator it = dts.end();
  --it;
 
  if (dm[i][j] < it->first) //把当前测试样本i到当前训练样本之间的欧氏距离与dts中最小距离比较，若更小就更新dts
  {
   dts.erase(it);
   dts.insert(make_pair(dm[i][j], train[j].type));
  }
  }
 }
 map<string, double> tds;
 string type = "";
 double weight = 0.0;
 //下面for循环主要是求出与测试样本i最邻近的k个样本点中大多数属于的类别，即将其作为测试样本点i的类别
 for (multimap<double, string>::const_iterator cit = dts.begin(); cit != dts.end(); ++cit)
 {
  // 不考虑权重的情况，在 k 个样例中只要出现就加 1
  // ++tds[cit->second];
 
  // 这里是考虑距离与权重的关系，距离越大权重越小
  tds[cit->second] += 1.0 / cit->first;
  if (tds[cit->second] > weight)
  {
  weight = tds[cit->second];
  type = cit->second; //保存一下类别
  }
 }
 test[i].type = type;
 }
}
 
// 计算欧氏距离
double euclideanDistance(const vector<double>& v1, const vector<double>& v2)
{
 if(v1.size() != v2.size())
 {
 cerr<<"Unable to get a distance! "<<endl;
 }
 
 else
 {
 double distance = 0.0;
 
 for (vector<double>::size_type i = 0; i != v1.size(); ++i)
 {
  distance += (v1[i] - v2[i]) * (v1[i] - v2[i]);
 }
 return sqrt(distance);
 }
}
 
/*初始化距离矩阵，该矩阵是根据训练样本和测试样本而得，
矩阵的行数为测试样本的数目，列数为训练样本的数目，
每一行为一个测试样本到各个训练样本之间的欧式距离组成的数组*/
void initDistanceMatrix(vector<vector<double> >& dm, const vector<sample>& train, const vector<sample>& test)
{
 for (vector<sample>::size_type i = 0; i != test.size(); ++i)
 {
 vector<double> vd;
 for (vector<sample>::size_type j = 0; j != train.size(); ++j)
 {
  vd.push_back(euclideanDistance(test[i].features, train[j].features));
 }
 dm.push_back(vd);
 }
}
 
//封装
void xfxKnn(const string& file1, const string& file2, const string& file3, int k)
{
 vector<sample> train,test;
 readTrain(train, file1.c_str());
 readTest(test, file2.c_str());
 vector< vector<double> > dm;
 initDistanceMatrix(dm, train, test);
 knnProcess(test, train, dm, k);
 writeResult(test, file3.c_str());
}
 
// 测试
int main()
{
 xfxKnn("train.txt", "test.txt", "result.txt", 5);
 return 0;
}