图分割算法之贪心算法

1 贪心算法的思想

        Linear Deterministic Greedy partitioning (LDG)考虑在分割的时候将邻居结点放置在一起，以减少切割边。它采用贪心算法将一个结点放置在包含其邻居最多的子图中，同时保证每个子图的结点负载均衡，整个算法流程图如下其中 C 表示每个分区的期望值，w(i)  表示当前子图在平衡状态下剩余容量，g(v,Pi)  表示再考虑负载的情况下结点 v 和子图 Pi 中结点邻居个数的交集，该打分函数作为将结点 v 分配到最大分数的子图中

2 代码设计

父类设计Partitioner

#ifndef PARTITION_H
#define PARTITION_H
#include <vector>
#include <algorithm>
#include <numeric>
#include <random>

class Partitioner 
{
public:
	/// <summary>
	/// 图分区的构造函数
	/// </summary>
	/// <param name="adjMatrix_">临界矩阵</param>
	Partitioner(std::vector<std::vector<int>> adjMatrix_);
	
	/// <summary>
	/// 图分割算法
	/// </summary>
	/// <param name="partNums">分区个数</param>
	virtual void execute(int partNums) = 0;

	/// <summary>
	/// 评估图分割算法
	/// </summary>
	virtual void evaluate() = 0;

	/// <summary>
	/// 返回分区结果
	/// </summary>
	std::vector<int> getResults() const;

	/// <summary>
	/// 获取顶点总数
	/// </summary>
	int getNumVertics() const;

	/// <summary>
	/// 获取图的便的总数
	/// </summary>
	int getNumEdges() const;

private:

protected:

	/// <summary>
	/// 邻接表存储的图数据
	/// </summary>
	std::vector<std::vector<int>> adjMatrix;
	/// <summary>
	/// 分区结果
	/// </summary>
	std::vector<int> partResults;

	int numVertices;

	int numEdges;

};

#endif // !PARTITION_H


#include "Partitioner.h"

Partitioner::Partitioner(std::vector<std::vector<int>> adjMatrix_) : adjMatrix(adjMatrix_)
{
    numVertices = adjMatrix_.size();
    numEdges = 0;
    std::for_each(adjMatrix_.begin(), adjMatrix_.end(), [&](const std::vector<int>& edges) {
        numEdges += edges.size();
        });
}

std::vector<int> Partitioner::getResults() const
{
    return partResults;
}

int Partitioner::getNumVertics() const
{
    return numVertices;
}

int Partitioner::getNumEdges() const
{
    return numEdges;
}

LDGPartitioner的设计

#ifndef LDG_PARTITIONER_H
#define LDG_PARTITIONER_H

#include "Partitioner.h"
#include <unordered_set>

class LDGPartitioner : public Partitioner
{
public:

	LDGPartitioner(std::vector<std::vector<int>> adjMatrix_) : Partitioner(adjMatrix_) {};

	void execute(int partNums) override;

	void evaluate() override;


private:
	/// <summary>
	/// 计算节点index的邻居和已分配的邻居相交的元素个数
	/// </summary>
	/// <param name="index">节点的下表</param>
	/// <param name="curVec">已分配的邻居</param>
	/// <returns></returns>
	int intersect(const int index, const std::unordered_set<int>& curVec);
	/// <summary>
	/// 已分配的集合
	/// </summary>
	std::vector<std::unordered_set<int>> curParts;
};

#endif // !LDG_PARTITIONER_H

#include "LDGPartitioner.h"

void LDGPartitioner::execute(int partNums)
{
	//初始化
	std::vector<int> order(numVertices); //节点id的集合
	curParts.resize(numVertices);
	partResults.resize(numVertices);

	// 随机打乱id
	std::iota(order.begin(), order.end(), 0);
	std::random_shuffle(order.begin(),order.end());

	//将前partNums的节点分配给每个分区作为第一个元素
	for (int i = 0; i < partNums; i++)
	{
		curParts[i].insert(order[i]);
		partResults[order[i]] = i;
	}
	// 每个分区的期望值
	double expectant = static_cast<double>(numVertices / partNums);

	//便利剩余的元素
	for (int ii = partNums; ii < numVertices; ii++)
	{
		//当前节点的id
		int vertex = order[ii];
		// 每个分区的得分
		std::vector<double> scores(partNums, 0);

		for (int jj = 0; jj < partNums; jj++) 
		{
			double curSize = curParts[jj].size();
			double weight = static_cast<double>(1 - (curSize / expectant));
			double neighbors = intersect(vertex, curParts[jj]);
			scores[jj] = neighbors * weight;
		}
		//节点需要分配给得分最高的节点
		int maxIndex = std::distance(scores.begin(), std::max_element(scores.begin(), scores.end()));
		curParts[maxIndex].insert(vertex);
		partResults[vertex] = maxIndex;

	}
}

void LDGPartitioner::evaluate()
{
}

int LDGPartitioner::intersect(const int index,const std::unordered_set<int>& curVec)
{
	int count = 0;

	for (const auto& element : adjMatrix[index])
	{
		if (curVec.count(element))
			count++;
	}
	return count;
}

主函数测试：

#include <iostream>
#include <vector>
#include <list>
#include <algorithm>


#include <memory>

#include "LDGPartitioner.h"

using namespace std;

// 定义图的邻接表类型
typedef vector<vector<int>> AdjacencyList;


int main() {
    // 构造示例图的邻接表
    AdjacencyList graph = {
        {1, 2, 4},
        {0, 2, 4},
        {0, 1, 3},
        {2, 4},
        {0, 1, 3}
    };
    //Partitioner* ptr = new LDGPartitioner(graph);
    unique_ptr<Partitioner> ptr = make_unique<LDGPartitioner>(graph);

    ptr->execute(3);

    
    // 初始化划分
    vector<int> partition = ptr->getResults();



    // 输出划分结果
    for (int i = 0; i < partition.size(); i++) {
        cout << "Vertex " << i << " belongs to Partition " << partition[i] << endl;
    }

    return 0;
}

3 执行结果