【Algorithm】最短路径算法

1 前言

从某顶点出发,沿图的边到达另一顶点所经过的路径中,各边上权值之和最小的一条路径叫做最短路径。解决最短路的问题有以下算法

  1. Dijkstra 算法
  2. Bellman-Ford 算法
  3. SPFA 算法(Bellman-Ford 算法的优化算法)
  4. Floyd 算法等

给出一些名词解释:

  • 单源最短路径:从图中某个顶点到图其他各顶点的最短路径
  • Relax 松弛:这是最短路径算法里面最关键的一步(我也不清楚怎么解释,自行体会吧)
  • 邻接矩阵与邻接表:都是表示图的方法,很简单,自行搜索
  • 负权值环:最简单的负权值环就是,自己到自己的一个负权值边(构成了一个回环);如此以来不断地通过这个环(路径),那我岂不可以无限缩短距离,真是鬼才

本文没有详细介绍(写起来太麻烦,还不如直接看视频容易理解)但是有推荐视频或文章,这里主要有我的具体实现,仅供参考

最后推荐一个 Github 项目,这是我在找资料的时候看到的,里面的算法实现均为 JAVA 语言,根据他的实现,本文提供 C++ 的实现,保持尽量了几种算法风格的统一,如发现错误,请指正~

2 Dijkstra 算法

关于这个算法,推荐看一下这个 Video 讲解,我在这里也就不介绍了,主要看实现

关键词:

  • 单源最短路径
  • 贪心算法

缺点:

  • 图中有负权值边时往往不适用,但可以选择 Bellman-Ford 算法代替。注:有负权值边导致出现负权值环才会使 Dijkstra 失效,而 Bellman-Ford 可以定位到负权值环,并且解决这个问题,后面会提到。

2.1 邻接矩阵实现

效率分析,我们用 V 代表顶点数, E 代表边数

  • 时间复杂度 O ( V 2 ) O(V^2) O(V2)

C++ 实现,一般来说需要以下四个完成

  • graph 是由邻接矩阵表示的图,相当于一个二维数组
  • vistied 负责维护访问过的节点,没访问 false,访问了设置 true
  • dis 负责维护由源点到每个顶点的最短距离
  • path 保存的是最短路径,在这里我是维护了完整的路径,但实际上只需要维护前一个点就可以了
#include <vector>
#include <iostream>
#include <algorithm>
#include <iomanip>
#include <limits>

using namespace std;

#define POSITIVE_INF (numeric_limits<int>::max())  // <limits> 里定义的

void Dijkstra(vector<vector<double> > &graph, vector<double> &dis, vector<int> *path, const int vNum, const int vStart) {

    fill_n(dis.begin(), vNum, POSITIVE_INF);
    dis[vStart] = 0;

    bool *visited = new bool[vNum];
    fill_n(visited, vNum, false);

    for (int k = 0; k < vNum; k++) {
        int minNum = POSITIVE_INF, minIndex = -1;
        for (int i = 0; i < vNum; i++) { //找一个最小距离的顶点
            if (false == visited[i] && dis[i] < minNum) {
                minNum = dis[i];
                minIndex = i;
            }
        }
        visited[minIndex] = true;
        for (int j = 0; j < vNum; j++) {
            if (true == visited[j]) continue; //已经访问过的跳过
            if (graph[minIndex][j] == POSITIVE_INF) continue; //通过 minIndex 无法到达 j,跳过
            if (dis[j] > dis[minIndex] + graph[minIndex][j]) { //Relax 操作
                dis[j] = dis[minIndex] + graph[minIndex][j]; //更新最短距离
                path[j] = path[minIndex]; //更新路径
                path[j].push_back(minIndex);
            }
        }
    }
    delete[]visited;
}

void AddEdge(vector<vector<double> > &graph, int from, int to, double cost) {
	graph[from][to] = cost;
}

void PrintPath(vector<int>& path, int self) {
    for (int v : path) {
        cout << v << "->";
    }
    cout << self << endl;
}

void PrintMatrix(vector<vector<double> > &graph) {
    for (auto level : graph) {
        cout << "[";
        for (auto v : level) {
            if (v >= POSITIVE_INF) cout << setw(5) << left << "inf";
            else cout << fixed << setprecision(0) << setw(5) << left << v;
        }
        cout << "]\n";
    }
    cout << endl;
}

int main() {
    int vNum = 5, vStart = 0;
    vector<vector<double> > graph(vNum, vector<double>(vNum, POSITIVE_INF));
    vector<double> dis(vNum, POSITIVE_INF); //记录源点到其他点的最短距离
    vector<int>* path = new vector<int>[vNum]; //记录最短距离经过的路径

    for(int i = 0; i < vNum; i++) AddEdge(graph, i, i, 0);
    AddEdge(graph, 0, 1, 6);
    AddEdge(graph, 0, 3, 1);
    AddEdge(graph, 1, 0, 6);
    AddEdge(graph, 1, 2, 5);
    AddEdge(graph, 1, 3, 2);
    AddEdge(graph, 1, 4, 2);
    AddEdge(graph, 2, 1, 5);
    AddEdge(graph, 2, 4, 5);
    AddEdge(graph, 3, 0, 1);
    AddEdge(graph, 3, 1, 2);
    AddEdge(graph, 3, 4, 1);
    AddEdge(graph, 4, 1, 2);
    AddEdge(graph, 4, 2, 5);
    AddEdge(graph, 4, 3, 1);

    PrintMatrix(graph);
    
    Dijkstra(graph, dis, path, vNum, vStart);

    for (int i = 0; i < vNum; i++) {
        cout << "From " << vStart << " to " << i << " distance: " << setw(5) << dis[i];
        PrintPath(path[i], i);
    }

    delete[]path;
    system("pause");

    // From 0 to 0 distance: 0    0
    // From 0 to 1 distance: 3    0->3->1
    // From 0 to 2 distance: 7    0->3->4->2
    // From 0 to 3 distance: 1    0->3
    // From 0 to 4 distance: 2    0->3->4
}

2.2 邻接表实现

效率分析,我们用 V 代表顶点数, E 代表边数

  • 简单实现,时间复杂度 O ( E V ) O(EV) O(EV)
  • 优先队列(最小堆),时间复杂度 O ( E l o g V ) O(ElogV) O(ElogV)

C++ 实现,不懂看注释

#include <vector>
#include <iostream>
#include <iomanip>
#include <algorithm>
#include <limits>
#include <queue>

using namespace std;

#define POSITIVE_INF (numeric_limits<int>::max())  // <limits> 里定义的

struct Edge {
    int from;
    int to;
    double cost;
    Edge(int from, int to, double cost) :from(from), to(to), cost(cost) {}
};

typedef pair<double, int> Node; // first 为到该顶点的最短距离,second 为顶点编号,这个数据结构用于优先队列里排序用的

void AddEdge(vector<Edge> *graph, int from, int to, double cost) {
    graph[from].push_back(Edge(from, to, cost));
}

//使用优先队列实现,时间复杂度 O((ElogV)
void Dijkstra_PQ(vector<Edge> *graph, vector<double> &dis, vector<int> *path, const int vNum, const int vStart) {
    
    fill_n(dis.begin(), vNum, POSITIVE_INF);
    dis[vStart] = 0;

    priority_queue<Node, vector<Node>, greater<Node> > q; //距离小的先出队,比较的是 Node.frist
    q.push(Node(0, vStart));

    bool *visited = new bool[vNum]; //设置已访问过的数组
    fill_n(visited, vNum, false);

    while (!q.empty()) {
        Node node = q.top(); q.pop();
        visited[node.second] = true;
        if (dis[node.second] < node.first) continue; //队列里的这个节点最短距离的值旧了,跳过
        for (Edge edge : graph[node.second]) {
            if (visited[edge.to]) continue; //已经访问过的节点则跳过
            if (dis[edge.to] > dis[edge.from] + edge.cost) { //Relax 操作
                dis[edge.to] = dis[edge.from] + edge.cost; //更新 distance
                path[edge.to] = path[edge.from]; //更新路径
                path[edge.to].push_back(edge.from);
                q.push(Node(dis[edge.to], edge.to));
            }
        }
    }

    delete[]visited;
}

//简单实现,时间复杂度 O(EV)
void Dijkstra_Simple(vector<Edge> *graph, vector<double> &dis, vector<int> *path, const int vNum, const int vStart) {

    fill_n(dis.begin(), vNum, POSITIVE_INF);
    dis[vStart] = 0;

    bool *visited = new bool[vNum];
    fill_n(visited, vNum, false);

    for (int k = 0; k < vNum; k++) {
        int minNum = POSITIVE_INF, minIndex = -1;
        for (int i = 0; i < vNum; i++) { //找一个最小的
            if (false == visited[i] && dis[i] < minNum) {
                minNum = dis[i];
                minIndex = i;
            }
        }
        visited[minIndex] = true;
        for (Edge edge : graph[minIndex]) {
            if (visited[edge.to]) continue; //已经访问过的节点则跳过
            if (dis[edge.to] > dis[edge.from] + edge.cost) { //Relax 操作
                dis[edge.to] = dis[edge.from] + edge.cost; //更新 distance
                path[edge.to] = path[edge.from]; //更新路径
                path[edge.to].push_back(edge.from);
            }
        }
    }

    delete[]visited;
}

void PrintPath(vector<int>& path, int self) {
    for (int v : path) {
        cout << v << "->";
    }
    cout << self << endl;
}

int main() {
    int vNum = 5, vStart = 0; // vNum 顶点数量,vStart 起始点编号

    vector<Edge> *graph = new vector<Edge>[vNum]; //邻接表表示的图
    vector<double> dis(vNum, POSITIVE_INF); //最短距离
    vector<int> *path = new vector<int>[vNum]; //存储记录的最短路径

    AddEdge(graph, 0, 1, 6);
    AddEdge(graph, 0, 3, 1);
    AddEdge(graph, 1, 0, 6);
    AddEdge(graph, 1, 2, 5);
    AddEdge(graph, 1, 3, 2);
    AddEdge(graph, 1, 4, 2);
    AddEdge(graph, 2, 1, 5);
    AddEdge(graph, 2, 4, 5);
    AddEdge(graph, 3, 0, 1);
    AddEdge(graph, 3, 1, 2);
    AddEdge(graph, 3, 4, 1);
    AddEdge(graph, 4, 1, 2);
    AddEdge(graph, 4, 2, 5);
    AddEdge(graph, 4, 3, 1);

    // Dijkstra_PQ(graph, dis, path, vNum, vStart);
    Dijkstra_Simple(graph, dis, path, vNum, vStart);

    for (int i = 0; i < vNum; i++) {
        cout << "From " << vStart << " to " << i << " distance: " << setw(5) << left << dis[i];
        PrintPath(path[i], i);
    }

    delete[]graph;
    delete[]path;

    system("pause");

    // From 0 to 0 distance: 0    0
    // From 0 to 1 distance: 3    0->3->1
    // From 0 to 2 distance: 7    0->3->4->2
    // From 0 to 3 distance: 1    0->3
    // From 0 to 4 distance: 2    0->3->4
}

3 Bellman Ford 算法

推荐食用这个 Video 以及这篇 Blog,我这里简单讲一下怎么解决负权值的问题。按照视频里说的,事实上就是将核心代码再跑一遍,如果还能进行松弛操作,那证明这里出现了负权值环了,然后可以不断扩散出去,能松弛的就代表受到负权值环的影响。

关键词:

  • 单源最短路径
  • 解决负权值环问题

缺点:

  • 时间复杂度过高,但可进行优化,即后面会提到的 SPFA 算法

3.1 邻接矩阵实现

效率分析,我们用 V 代表顶点数, E 代表边数

  • 时间复杂度 O ( V 3 ) O(V^3) O(V3)
#include <vector>
#include <iomanip>
#include <iostream>
#include <queue>
#include <limits>

using namespace std;

#define POSITIVE_INF (numeric_limits<int>::max())
#define NEGATIVE_INF (numeric_limits<int>::min())

void BellmanFord(vector<vector<double> > &graph, vector<double> &dis, vector<int>* path ,const int vNum, const int vStart) {

    fill_n(dis.begin(), vNum, POSITIVE_INF);
    dis[vStart] = 0;

    for (int k = 0; k < vNum - 1; k++) { // For each vertex, apply relaxation for all the edges
        for (int i = 0; i < vNum; i++) {
            for (int j = 0; j < vNum; j++) {
                if (graph[i][j] == POSITIVE_INF) continue; // 没有 i 到 j 的边,跳过
                if (dis[j] > dis[i] + graph[i][j]) { // Relax 操作
                    dis[j] = dis[i] + graph[i][j]; // 更新最短距离
                    path[j] = path[i]; // 更新路径
                    path[j].push_back(i);
                }
            }   
        }
    }

    for (int k = 0; k < vNum - 1; k++) { // Run algorithm a second time to detect which nodes are part of a negative cycle.
        for (int i = 0; i < vNum; i++) {
            for (int j = 0; j < vNum; j++) {
                if (graph[i][j] == POSITIVE_INF) continue; // 没有 i 到 j 的边,跳过
                if (dis[j] == NEGATIVE_INF) continue;
                if (dis[j] > dis[i] + graph[i][j]) {
                    dis[j] = NEGATIVE_INF;
                    path[j].clear();
                }
            }
        }
    }
}

void AddEdge(vector<vector<double> > &graph, int from, int to, double cost) {
	graph[from][to] = cost;
}

void PrintPath(vector<int>& path, int self) {
    for (int v : path) {
        cout << v << "->";
    }
    cout << self << endl;
}

void PrintMatrix(vector<vector<double> > &graph) {
    for (auto level : graph) {
        cout << "[";
        for (auto v : level) {
            if (v >= POSITIVE_INF) cout << setw(5) << left << "inf";
            else cout << fixed << setprecision(0) << setw(5) << left << v;
        }
        cout << "]\n";
    }
    cout << endl;
}

int main() {
    int vNum = 9, vStart = 0;
	vector<vector<double> > graph(vNum, vector<double>(vNum, POSITIVE_INF));
    vector<double> dis(vNum, POSITIVE_INF);
    vector<int>* path = new vector<int>[vNum];

    for (int i = 0; i < vNum; i++) AddEdge(graph, i, i, 0);
    AddEdge(graph, 0, 1, 1);
    AddEdge(graph, 1, 2, 1);
    AddEdge(graph, 2, 4, 1);
    AddEdge(graph, 4, 3, -3);
    AddEdge(graph, 3, 2, 1);
    AddEdge(graph, 1, 5, 4);
    AddEdge(graph, 1, 6, 4);
    AddEdge(graph, 5, 6, 5);
    AddEdge(graph, 6, 7, 4);
    AddEdge(graph, 5, 7, 3);

    PrintMatrix(graph);

    BellmanFord(graph, dis, path, vNum, vStart);

    for (int i = 0; i < vNum; i++) {
        cout << "From " << vStart << " to " << i << " distance: ";
        if (dis[i] == POSITIVE_INF) cout << left << setw(6) << "+inf";
        else if (dis[i] == NEGATIVE_INF) cout << left << setw(6) << "-inf";
        else cout << left << setw(6) << dis[i];
        PrintPath(path[i], i);
    }

    delete[]path;
    
    system("pause");
}

3.2 邻接表实现

效率分析,我们用 V 代表顶点数, E 代表边数

  • 时间复杂度 O ( E V ) O(EV) O(EV)
#include <vector>
#include <iostream>
#include <iomanip>
#include <algorithm>
#include <limits>

using namespace std;

#define POSITIVE_INF (numeric_limits<int>::max())
#define NEGATIVE_INF (numeric_limits<int>::min())

struct Edge {
    int from;
    int to;
    double cost;
    Edge(int from, int to, double cost) :from(from), to(to), cost(cost) {}
};

void AddEdge(vector<Edge>* graph, int from, int to, double cost) {
    graph[from].push_back(Edge(from, to, cost));
}

void BellmanFord(vector<Edge>* graph, vector<double> &dis, vector<int>* path, const int vNum, const int vStart) {

    fill_n(dis.begin(), vNum, POSITIVE_INF);
    dis[vStart] = 0;

    for (int k = 0; k < vNum - 1; k++) {  // For each vertex, apply relaxation for all the edges
        for (int i = 0; i < vNum; i++) {
            for (Edge edge : graph[i]) {
                if (dis[edge.from] + edge.cost < dis[edge.to]) {
                    dis[edge.to] = dis[edge.from] + edge.cost;
                    path[edge.to] = path[edge.from];
                    path[edge.to].push_back(edge.from);
                }
            }
        }
    }
    for (int k = 0; k < vNum - 1; k++) { // Run algorithm a second time to detect which nodes are part of a negative cycle. 
        for (int i = 0; i < vNum; i++) { 
            for (Edge edge : graph[i]) {
                if (dis[edge.from] + edge.cost < dis[edge.to]) {
                    dis[edge.to] = NEGATIVE_INF;
                    path[edge.to].clear();
                }
            }
        }
    }
}

void PrintPath(vector<int>& path, int self) {
    for (int v : path) {
        cout << v << "->";
    }
    cout << self << endl;
}

void PrintEdge(vector<Edge> *graph, int vNum) {
    for (int i = 0; i < vNum; i++) {
        for (Edge edge : graph[i]) {
            cout << "Edge " << edge.from << " to " << edge.to << " , cost: " << edge.cost << endl;
        }
    }
    cout << endl;
}

int main() {
    int vNum = 9, vStart = 0;
    vector<Edge> *graph = new vector<Edge>[vNum];
    vector<double> dis(vNum, POSITIVE_INF);
    vector<int>* path = new vector<int>[vNum];

    AddEdge(graph, 0, 1, 1);
    AddEdge(graph, 1, 2, 1);
    AddEdge(graph, 2, 4, 1);
    AddEdge(graph, 4, 3, -3);
    AddEdge(graph, 3, 2, 1);
    AddEdge(graph, 1, 5, 4);
    AddEdge(graph, 1, 6, 4);
    AddEdge(graph, 5, 6, 5);
    AddEdge(graph, 6, 7, 4);
    AddEdge(graph, 5, 7, 3);

    PrintEdge(graph, vNum);

    BellmanFord(graph, dis, path, vNum, vStart);

    for (int i = 0; i < vNum; i++) {
        cout << "From " << vStart << " to " << i << " distance: ";
        if (dis[i] == POSITIVE_INF) cout << left << setw(6) << "+inf";
        else if (dis[i] == NEGATIVE_INF) cout << left << setw(6) << "-inf";
        else cout << left << setw(6) << dis[i];
        PrintPath(path[i], i);
    }

    delete[]graph;
    delete[]path;

    system("pause");

    // [0    1    inf  inf  inf  inf  inf  inf  inf  ]
    // [inf  0    1    inf  inf  4    4    inf  inf  ]
    // [inf  inf  0    inf  1    inf  inf  inf  inf  ]
    // [inf  inf  1    0    inf  inf  inf  inf  inf  ]
    // [inf  inf  inf  -3   0    inf  inf  inf  inf  ]
    // [inf  inf  inf  inf  inf  0    5    3    inf  ]
    // [inf  inf  inf  inf  inf  inf  0    4    inf  ]
    // [inf  inf  inf  inf  inf  inf  inf  0    inf  ]
    // [inf  inf  inf  inf  inf  inf  inf  inf  0    ]

    // From 0 to 0 distance: 0     0
    // From 0 to 1 distance: 1     0->1
    // From 0 to 2 distance: -inf  2
    // From 0 to 3 distance: -inf  3
    // From 0 to 4 distance: -inf  4
    // From 0 to 5 distance: 5     0->1->5
    // From 0 to 6 distance: 5     0->1->6
    // From 0 to 7 distance: 8     0->1->5->7
    // From 0 to 8 distance: +inf  8
}

3.3 一些优化方法

优化一:循环的提前跳出

通常 Bellman Ford 算法经常会在未达到 n-1 轮松弛前就已经计算出最短路。因此每一轮执行后,若没有执行松弛操作则算法就可以停止了,则我们可以添加一个标志位 flag,如下所示

for (int k = 0; k < vNum - 1; k++) { // 一共是执行 vNum - 1 轮
        bool flag = false;
        for (int i = 0; i < vNum; i++) {
            for (Edge edge : graph[i]) {
                if (dis[edge.from] + edge.cost < dis[edge.to]) {
                    dis[edge.to] = dis[edge.from] + edge.cost;
                    path[edge.to] = path[edge.from];
                    path[edge.to].push_back(edge.from);
                    flag = true; // 更新了 dis 则标志设 true
                }
            }
        }
        if(!flag) break; // dis 未更新,跳出循环
    }

优化二:SPFA 的算法,即避免冗余的计算

松弛操作必定只会发生在最短路径前导节点松弛成功过的节点上,则用一个队列记录松弛过的节点,可以避免了冗余计算,这即是 SPFA 算法的核心,下一节将介绍 SPFA 的实现

4 SPFA 算法

通过前面我们知道 SPFA 算法是优化后的 Bellman Ford 算法,如何实现?以下引用自百度百科-SPFA算法

动态逼近法:设立一个先进先出的队列用来保存待优化的结点,优化时每次取出队首结点u,并且用u点当前的最短路径估计值对离开u点所指向的结点v进行松弛操作,如果v点的最短路径估计值有所调整,且v点不在当前的队列中,就将v点放入队尾

关于 SPFA 算法,我是仿照百度百科里实现的,里面检测到环的时候将会直接中止算法,为了与前面保持一致,我做了一些改进。与前面一样,改进之后能检测出哪些节点会受到负权值环的影响,受到影响的其最后输出的最短距离为 -inf

注:此代码可能有误,请大家小心参考

#include <vector>
#include <iostream>
#include <iomanip>
#include <algorithm>
#include <limits>
#include <queue>

using namespace std;

#define POSITIVE_INF (numeric_limits<int>::max())
#define NEGATIVE_INF (numeric_limits<int>::min())

struct Edge {
    int from;
    int to;
    double cost;
    Edge(int from, int to, double cost) :from(from), to(to), cost(cost) {}
};

void AddEdge(vector<Edge>* graph, int from, int to, double cost) {
	graph[from].push_back(Edge(from, to, cost));
}

void SPFA(vector<Edge>* graph, vector<double> &dis, vector<int>* path, const int vNum, const int vStart) {

    fill_n(dis.begin(), vNum, POSITIVE_INF);
    dis[vStart] = 0;

    queue<int> q;
    q.push(vStart);

    vector<bool> inQueue(vNum, false); //记录某顶点是否在队列中
    inQueue[vStart] = true;

    vector<int> cnt(vNum, 0); //记录某顶点进入队列的次数
    cnt[vStart] = 1;
    
    while (!q.empty()) {
        int v = q.front(); q.pop();
        inQueue[v] = false;
        for (Edge edge : graph[v]) {
            if (dis[edge.to] == NEGATIVE_INF) continue; //已经被负权环路感染
            if (dis[edge.from] == NEGATIVE_INF) { //将要被负权环路感染
                dis[edge.to] = NEGATIVE_INF;
                path[edge.to].clear();
                if(!inQueue[edge.to]) {
                    q.push(edge.to);
                    inQueue[edge.to] = true;
                }
                continue;
            }
            if (dis[edge.to] > dis[edge.from] + edge.cost) { //松弛
                dis[edge.to] = dis[edge.from] + edge.cost;
                path[edge.to] = path[edge.from];
                path[edge.to].push_back(edge.from);
                if (!inQueue[edge.to]) {
                    cnt[edge.to]++;
                    if (cnt[edge.to] == vNum) { //出现负权值
                        dis[edge.to] = NEGATIVE_INF;
                        path[edge.to].clear();
                    }
                    q.push(edge.to);
                    inQueue[edge.to] = true;
                }
            }
        }
    }
}

void PrintPath(vector<int>& path, int self) {
    for (int v : path) {
        cout << v << "->";
    }
    cout << self << endl;
}

void PrintEdge(vector<Edge> *graph, int vNum) {
    for (int i = 0; i < vNum; i++) {
        for (Edge edge : graph[i]) {
            cout << "Edge " << edge.from << " to " << edge.to << " , cost: " << edge.cost << endl;
        }
    }
    cout << endl;
}

int main() {
	int vNum = 9, vStart = 0;
	vector<Edge> *graph = new vector<Edge>[vNum];
    vector<double> dis(vNum, POSITIVE_INF);
    vector<int>* path = new vector<int>[vNum];

    AddEdge(graph, 0, 1, 1);
    AddEdge(graph, 1, 2, 1);
    AddEdge(graph, 2, 4, 1);
    AddEdge(graph, 4, 3, -3);
    AddEdge(graph, 3, 2, 1);
    AddEdge(graph, 1, 5, 4);
    AddEdge(graph, 1, 6, 4);
    AddEdge(graph, 5, 6, 5);
    AddEdge(graph, 6, 7, 4);
    AddEdge(graph, 5, 7, 3);

    PrintEdge(graph, vNum);

    SPFA(graph, dis, path, vNum, vStart);

    for (int i = 0; i < vNum; i++) {
        cout << "From " << vStart << " to " << i << " distance: ";
        if (dis[i] == POSITIVE_INF) cout << left << setw(6) << "+inf";
        else if (dis[i] == NEGATIVE_INF) cout << left << setw(6) << "-inf";
        else cout << left << setw(6) << dis[i];
        PrintPath(path[i], i);
    }

    delete[]graph;
    delete[]path;

    system("pause");
}

5 Floyd-Warshall 算法

先推荐 Video,这是本文实现的最后一个算法,这也是大家一般学习到的多源最短路径算法,需要注意的是

  • dis 数组上升到了二维(毕竟是多源最短路径)
  • 保存路径的不再是 pre 数组,而是 next 数组,看完视频后好好想想他们之间的区别

5.1 邻接矩阵实现

#include <vector>
#include <iostream>
#include <algorithm>
#include <iomanip>
#include <limits>

using namespace std;

#define POSITIVE_INF (numeric_limits<int>::max())
#define NEGATIVE_INF (numeric_limits<int>::min())
#define DOES_NOT_EXIST -1
#define REACHES_NEGATIVE_CYCLE -2

void AddEdge(vector<vector<double> > &graph, int from, int to, double cost) {
	graph[from][to] = cost;
}

void Floyd(vector<vector<double> > &graph, vector<vector<double> > &dis, vector<vector<int> > &next, const int vNum) {

    dis = graph; //赋值操作,不是 java 里的指向对象哈
    
    for (int i = 0; i < vNum; i++) { //pre 数组初始化
        for (int j = 0; j < vNum; j++) {
            if (graph[i][j] != POSITIVE_INF) next[i][j] = j;
        }
    }

    for (int k = 0; k < vNum; k++) {
        for (int i = 0; i < vNum; i++) {
            for (int j = 0; j < vNum; j++) {
                if (dis[i][k] == POSITIVE_INF || dis[k][j] == POSITIVE_INF) continue;
                if (dis[i][j] > dis[i][k] + dis[k][j]) {
                    dis[i][j] = dis[i][k] + dis[k][j];
                    next[i][j] = next[i][k];
                }
            }
        }
    }

    // Identify negative cycles by propagating the value 'NEGATIVE_INFINITY'
    // to every edge that is part of or reaches into a negative cycle.
    for (int k = 0; k < vNum; k++) {
        for (int i = 0; i < vNum; i++) {
            for (int j = 0; j < vNum; j++) {
                if (dis[i][k] == POSITIVE_INF || dis[k][j] == POSITIVE_INF) continue;
                if (dis[i][j] == NEGATIVE_INF) continue;
                if (dis[i][j] > dis[i][k] + dis[k][j]) {
                    dis[i][j] = NEGATIVE_INF;
                    next[i][j] = REACHES_NEGATIVE_CYCLE;
                }
            }
        }
    }
}

template<typename T>
void PrintMatrix(vector<vector<T> > &graph) {
    for (auto level : graph) {
        cout << "[";
        for (auto v : level) {
            if (v >= POSITIVE_INF) cout << setw(5) << left << "+inf";
            else if ( v <= NEGATIVE_INF) cout << setw(5) << left << "-inf";
            else cout << fixed << setprecision(0) << setw(5) << left << v;
        }
        cout << "]\n";
    }
    cout << endl;
}

int main() {
    int vNum = 7;
    vector<vector<double> > graph(vNum, vector<double>(vNum, POSITIVE_INF)); //图中所有值初始化为 POSITIVE_INF 即两点不联通
    vector<vector<double> > dis(vNum, vector<double>(vNum, POSITIVE_INF)); //dis[i][j] 记录当前 i 到 j 的最短距离,是动态规划中的 dp 数组
    vector<vector<int> > next(vNum, vector<int>(vNum, DOES_NOT_EXIST)); //保存 i 到 j 当前最短路径的下一个节点

    //设置图
    for (int i = 0; i < vNum; i++) graph[i][i] = 0;
    AddEdge(graph, 0, 1, 2);
    AddEdge(graph, 0, 2, 5);
    AddEdge(graph, 0, 6, 10);
    AddEdge(graph, 1, 2, 2);
    AddEdge(graph, 1, 4, 11);
    AddEdge(graph, 2, 6, 2);
    AddEdge(graph, 6, 5, 11);
    AddEdge(graph, 4, 5, 1);
    AddEdge(graph, 5, 4, -2);

    PrintMatrix(graph);

    Floyd(graph, dis, next, vNum);
    
    PrintMatrix(dis);
    PrintMatrix(next);

    system("pause");
}

6 最后我想说的

我的实现并不是标准,仅供大家参考,如果你有更好的实现也欢迎在评论区留言。如果有误请立即联系我,感谢。

然后,本文只列举了 4 种算法,还有其他的并未提及,所以可能还会有续篇~

7 参考资料

[1] 百度百科/Bellman-Ford算法 https://baike.baidu.com/item/Bellman-Ford算法

[2] Bellman-Ford算法(解决负权边)https://blog.csdn.net/wtyvhreal/article/details/43450727

[3] SPFA算法 https://baike.baidu.com/item/SPFA%E7%AE%97%E6%B3%95

[4] Github/williamfiset https://github.com/williamfiset/Algorithms

[5] 最短路径搜索之A*算法 https://blog.csdn.net/yuxuan20062007/article/details/80914210

评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值