二分K均值c++实现

二分k均值原理及Python实现请见：

（1）http://blog.youkuaiyun.com/u013593585/article/details/51280052

（2）http://blog.youkuaiyun.com/u013593585/article/details/51263980 Python实现

C++代码：基于python的思路写的(求指点)

#include <iostream>
#include <fstream>
#include <sstream>
#include <vector>
#include <ctime>
using namespace std;
typedef vector<float> Tuple;
const int dims = 2;
const int k = 2;
const int bik = 4;

void doKmeans(vector<Tuple>& tuples,Tuple subcentroids[],vector<int>& SubclusterAssment1,Tuple& SubclusterAssment2);  
void assignTuples(vector<Tuple> clusters[],vector<Tuple> tuples,Tuple means[]);  
float getDist(const Tuple& t1, const Tuple& t2);  
float getVal(vector<Tuple> clusters[], Tuple means[]);  
Tuple updateMeans(const vector<Tuple>& cluster_i);  
void print(vector<Tuple> clustes[]); 

int main()
{
	//open the file
	char filename[] = "testSet.txt";
	fstream file(filename);
	if(!file)
	{
		cout << "can not opnen the file!" << endl;
		return 0;
	}

	//read the data读取数据
	vector<Tuple> tuples;

	int pos = 0;
	while(!file.eof())
	{
		string str;
		getline(file,str);
		stringstream ss(str);
		Tuple tuple(dims+1,0); //the first element in tuple indicates the index
		tuple[0] = pos + 1;
		for(int i = 1; i <= dims; i++)
			ss >> tuple[i];
		tuples.push_back(tuple);
	}

	vector<Tuple> centroids; // generated centroids 生成的聚类中心
	vector<int> clusterAssment1;  // samples index in different centroids 样本对应的簇索引
	Tuple clusterAssment2;  // samples se 每个样本到其聚类中心的平方误差
	
	//initialize the first centroid初始化第一个聚类中心（取样本均值）
	Tuple centroid(dims+1,0);
	for(int i = 0; i < tuples.size();i++)
	{
		for(int j = 1; j <= dims; j++)
			centroid[j] = centroid[j] + tuples[i][j];
	}
	for(int j = 1; j <= dims; j++)
		centroid[j] /= tuples.size();
	centroids.push_back(centroid);
	//initialize the clusterAssment1and clusterAssmen2 初始化样本簇索引为0及计算样本到聚类中心的距离</span>
	for(int i = 0; i < tuples.size(); i++)
	{
		clusterAssment1.push_back(0);
		clusterAssment2.push_back(getDist(tuples[i],centroid));
	}

	//inviding二分
	int test = 0;
	while(centroids.size() < bik) //是否已经得到bik个聚类中心
	{
		cout << "centroids.size: " << centroids.size() << endl;
		//compute the SSE of each centroid in centroids  对于每个簇计算其SSE平方和误差 及 对该簇二聚类下的 样本总聚类代价很熟，</span><span style="font-family:Arial, Helvetica, sans-serif;">进而选择对哪个簇进行二聚类？
		Tuple minSubcentroid[2]; //最终选择的簇的两个子簇的聚类中心
		vector<int> minSubclusterAssment1; //最终选择的簇的两个子簇的聚类中心的簇索引
		Tuple minSubclusterAssment2;   //最终选择的簇的两个子簇到各自聚类中心的距离
		int minIndex; //最终选择的簇索引
		int minSSE = 10000;
		//for each cluster i 对于每个聚类中心进行操作
		for(int i = 0; i < centroids.size(); i++)
		{
			//find the subdataset belonging to cluster i //记录属于该聚类中心的样本
			vector<Tuple> SubdataSet;
			for(int j = 0; j < tuples.size();j++)
			{
				if (clusterAssment1[j] == i)
					SubdataSet.push_back(tuples[j]);
			}
			//2means clustering //对该簇进行二聚类
			Tuple Subcentroids[2];
			vector<int> SubclusterAssment1;
			Tuple SubclusterAssment2;
			doKmeans(SubdataSet,Subcentroids,SubclusterAssment1,SubclusterAssment2);
			//computer the sum SSE of cluster i //二聚类后，计算对应SSE 及总的聚类代价 
			float SSEi = 0;
			float SSEother = 0;
			float SSEsum = 0
                        //该簇聚类后的SSE
			for (int x = 0; x < SubclusterAssment2.size(); x++)
				SSEi += SubclusterAssment2[x];
                        //其他簇的SSE
			for (int y = 0; y < tuples.size(); y++)
			{
				if (clusterAssment1[y] != i)
					SSEother += clusterAssment2[y];
			}
			SSEsum = SSEi + SSEother;
			//find the cluster who has the minimal SSE
			if( SSEsum < minSSE) 
			{
				minSubcentroid[0] = Subcentroids[0];
				minSubcentroid[1] = Subcentroids[1];
				for (int ii = 0; ii < SubclusterAssment1.size(); ii++)
				{
					minSubclusterAssment1.push_back(SubclusterAssment1[ii]);
					minSubclusterAssment2.push_back(SubclusterAssment2[ii]);
				}
				minIndex = i; 
				minSSE = SSEsum;
			}
		}
		//update the centroids  <span style="font-family: Arial, Helvetica, sans-serif;">//更新聚类中心 </span>
		int yuan_Size = centroids.size();
		centroids[minIndex].swap(minSubcentroid[0]); //minIndex簇二聚类后索引为0的子簇的聚类中心取代该簇原来的聚类中心，
		centroids.push_back(minSubcentroid[1]); //minIndex簇二聚类后索引为1的子簇的聚类中心添加到后面
		cout << "yuan minSubclusterAssment1: " << endl;
		for(int jj = 0; jj < minSubclusterAssment1.size(); jj++)
				cout << minSubclusterAssment1[jj] << " ";
		cout << endl;
		//update the clusterAssment //更新样本的簇索引
                           //更新两个新簇的索引
		for(int s = 0; s < minSubclusterAssment1.size(); s++)
		{
			int temp = minSubclusterAssment1[s];
			if (temp == 1)
				minSubclusterAssment1[s] = yuan_Size;
			else
				minSubclusterAssment1[s] = minIndex;
		}
		cout << "minIndex: " << minIndex << endl;
		cout << "updated minSubclusterAssment1: " << endl;
		for(int jj = 0; jj < minSubclusterAssment1.size(); jj++)
				cout << minSubclusterAssment1[jj] << " ";
		cout << endl;
                          //更新样本簇索引，将样本簇索引为minIndex的全部换成两个新子簇的索引
		int count = 0;
		for(int a = 0; a < clusterAssment1.size(); a++)
		{
			if(clusterAssment1[a] == minIndex)
			{
				clusterAssment1[a] = minSubclusterAssment1[count];
				count++;
			}
		}
		cout << "clusterAssment1: " << endl;
		for(int jj = 0; jj < clusterAssment1.size(); jj++)
			cout << clusterAssment1[jj] << " ";
		cout << endl;

	}
	cout << "biKmenas is done!" << endl;
	system("pause");
	return 0;
}

void doKmeans(vector<Tuple>& tuples,Tuple subcentroids[],vector<int>& SubclusterAssment1,Tuple& SubclusterAssment2)  
{  
	vector<Tuple> clusters[k];  
    Tuple means[k];  
    srand((unsigned)time(NULL));  
    for (int i = 0; i < k; i++)  
    {  
        int temp = rand()%tuples.size();  
        means[i] = tuples[temp];   
    }  
    assignTuples(clusters,tuples,means);  
    double newVal = getVal(clusters,means);  
    double oldVal = -1;  
    int t = 0;  
    while ((abs(newVal - oldVal)) > 1)  
    {  
        for (int i = 0; i < k; i++)  
            means[i] = updateMeans(clusters[i]);  
        oldVal = newVal;  
        newVal = getVal(clusters,means);  
    }  
	subcentroids[0] = means[0];
	subcentroids[1] = means[1];
	for(int i = 0; i < tuples.size();i++)
	{
        double dist0 = getDist(tuples[i],means[0]);  
        double dist1 = getDist(tuples[i],means[1]); 
		int label = 0; 
		int dist = dist0;
        if (dist1 < dist0)  
        {  
            dist = dist1;  
            label = 1;  
        }  
		SubclusterAssment1.push_back(label);
		SubclusterAssment2.push_back(dist);
	}
 
}  
void assignTuples(vector<Tuple> clusters[], vector<Tuple> tuples,Tuple means[])  
{  
    for (int i = 0; i < tuples.size(); i++)  
    {  
        int label = 0;  
        double dist = getDist(tuples[i],means[0]);  
        for (int j = 1; j < k; j++)  
        {  
            double temp = getDist(tuples[i],means[j]);  
            if (temp < dist)  
            {  
                dist = temp;  
                label = j;  
            }  
        }  
        clusters[label].push_back(tuples[i]);  
    }  
}  
  
float getDist(const Tuple& t1, const Tuple& t2)  
{  
	float sum = 0;  
    for (int i = 1; i <= dims; i++)  
        sum += (t1[i] - t2[i]) * (t1[i] - t2[i]);  
    return sum;  
}  
  
float getVal(vector<Tuple> clusters[], Tuple means[])  
{  
    float val = 0;  
    for (int i = 0; i < k; i++)  
    {  
        vector<Tuple> t = clusters[i];  
        for (int j = 0; j < t.size(); j++)  
            val += getDist(t[j],means[i]);  
    }  
    return val;  
}  
  
Tuple updateMeans(const vector<Tuple>& cluster_i)  
{  
    Tuple t(dims+1,0);  
    for (int i = 0; i < cluster_i.size(); i++)  
        for (int j = 1; j <= dims; j++)  
            t[j] += cluster_i[i][j];  
    for (int i = 1; i <= dims; i++)  
        t[i] /= cluster_i.size();  
    return t;  
}  

实现结果：