目录
一.图的基本概念
图是由顶点集合及顶点间的关系组成的一种数据结构:G = (V, E),其中:
顶点集合V = {x|x属于某个数据对象集}是有穷非空集合;
E = {(x,y)|x,y属于V}或者E = {|x,y属于V && Path(x, y)}是顶点间关系的有穷集合,也叫做边的集合
(x, y)表示x到y的一条双向通路,即(x, y)是无方向的;Path(x, y)表示从x到y的一条单向通路,即Path(x, y)是有方向的
顶点和边:图中结点称为顶点 ,第i个顶点记作vi。两个顶点vi和vj相关联称作顶点vi和顶点vj之间有一条边,图中的第k条边记作ek,ek = (vi,vj)或<vi,vj>
有向图和无向图:在有向图中,顶点对是有序的,顶点对<x,y>称为顶点x到顶点y的一条边(弧),和是两条不同的边 ,比如下图G3和G4为有向图。在无向图中,顶点对(x, y) 是无序的,顶点对(x,y)称为顶点x和顶点y相关联的一条边,这条边没有特定方向,(x, y)和(y,x) 是同一条边,比如下图G1和G2为无向图。注意:无向边(x, y)等于有向边和
完全图:在有n个顶点的无向图中,若有n * (n-1)/2条边 ,即任意两个顶点之间有且仅有一条边, 则称此图为无向完全图 ,比如上图G1;在n个顶点的有向图中,若有n * (n-1)条边 ,即任意两个顶点之间有且仅有方向相反的边,则称此图为有向完全图,比如上图G4
邻接顶点:在无向图中G中,若(u, v)是E(G)中的一条边,则称u和v互为邻接顶点 ,并称边(u,v)依附于顶点u和v;在有向图G中,若是E(G)中的一条边,则称顶点u邻接到v,顶点v邻接自顶 点u,并称边与顶点u和顶点v相关联
顶点的度:顶点v的度是指与它相关联的边的条数,记作deg(v)。在有向图中,顶点的度等于该顶点的入度与出度之和, 其中顶点v的入度是以v为终点的有向边的条数,记作indev(v);顶点v的出度是以v为起始点的有向边的条数,记作outdev(v)。因此:dev(v) = indev(v) + outdev(v) 。注 意:对于无向图,顶点的度等于该顶点的入度和出度 ,即dev(v) = indev(v) = outdev(v)
路径: 在图G = (V, E)中,若从顶点vi出发有一组边使其可到达顶点vj,则称顶点vi到顶点vj的顶点序列为从顶点vi到顶点vj的路径
路径长度:对于不带权的图,一条路径的路径长度是指该路径上的边的条数;对于带权的图,一 条路径的路径长度是指该路径上各个边权值的总和
简单路径与回路:若路径上各顶点v1,v2,v3,...,vm均不重复,则称这样的路径为简单路 径。若路径上第一个顶点v1和最后一个顶点vm重合,则称这样的路径为回路或环
子图:设图G = {V, E}和图G1 = {V1,E1},若V1属于V且E1属于E,则称G1是G的子图
连通图:在无向图中,若从顶点v1到顶点v2有路径,则称顶点v1与顶点v2是连通的。如果图中任意一 对顶点都是连通的,则称此图为连通图
强连通图:在有向图中,若在每一对顶点vi和vj之间都存在一条从vi到vj的路径,也存在一条从vj到vi的路径,则称此图是强连通图
生成树:一个连通图的最小连通子图称作该图的生成树。有n个顶点的连通图的生成树有n个顶点和n1条边
二.图的存储结构
1.邻接矩阵
因为节点与节点之间的关系就是连通与否,即为0或者1,因此邻接矩阵(二维数组)即是:先用一 个数组将定点保存,然后采用矩阵来表示节点与节点之间的关系
注意:
- 无向图的邻接矩阵是对称的, 第i行(列)元素之和,就是顶点i的度。**有向图的邻接矩阵则不一 定是对称的,**第i行(列)元素之后就是顶点i的出(入)度
- 如果边带有权值,并且两个节点之间是连通的,上图中的边的关系就用权值代替,如果两个 顶点不通,则使用无穷大代替
- 用邻接矩阵存储图的有点是能够快速知道两个顶点是否连通,缺陷是如果顶点比较多,边比 较少时,矩阵中存储了大量的0成为系数矩阵,比较浪费空间,并且要求两个节点之间的路 径不是很好求
cpp
namespace matrix
{
template<class V, class W, W MAX_W = INT_MAX, bool Direction = false>
class Graph
{
typedef Graph<V, W, MAX_W, Direction> Self;
public:
//图的创建
//1.IO输入->不方便测试,oj中更适合
//2.图结构关系写到文件,读取文件
//3.手动添加边
Graph() = default;
Graph(const V* a, size_t n)
{
_vertexs.reserve(n);
for (size_t i = 0; i < n; ++i)
{
_vertexs.push_back(a[i]);
_indexMap[a[i]] = i;
}
_matrix.resize(n);
for (size_t i = 0; i < _matrix.size(); ++i)
{
_matrix[i].resize(n, MAX_W);
}
}
size_t GetVertexIndex(const V& v)
{
auto it = _indexMap.find(v);
if (it != _indexMap.end())
{
return it->second;
}
else
{
//assert(false);
throw invalid_argument("顶点不存在");
return -1;
}
}
void _AddEdge(size_t srci, size_t dsti, const W& w)
{
_matrix[srci][dsti] = w;
//无向图
if (Direction == false)
{
_matrix[dsti][srci] = w;
}
}
void AddEdge(const V& src, const V& dst, const W& w)
{
size_t srci = GetVertexIndex(src);
size_t dsti = GetVertexIndex(dst);
_AddEdge(srci, dsti, w);
}
void Print()
{
//顶点
for (size_t i = 0; i < _vertexs.size(); ++i)
{
cout << "[" << i << "]" << "->" << _vertexs[i] << endl;
}
cout << endl;
//矩阵
//横下标
cout << " ";
for (size_t i = 0; i < _matrix.size(); ++i)
{
//cout << i << " ";
printf("%4d", i);
}
cout << endl;
for (size_t i = 0; i < _matrix.size(); ++i)
{
cout << i << " ";//竖下标
for (size_t j = 0; j < _matrix[i].size(); ++j)
{
//cout << _matrix[i][j] << " ";
if (_matrix[i][j] == MAX_W)
{
//cout << "* ";
printf("%4c", '*');
}
else
{
//cout << _matrix[i][j] << " ";
printf("%4d", _matrix[i][j]);
}
}
cout << endl;
}
cout << endl;
}
private:
vector<V> _vertexs; //顶点集合
map<V, int> _indexMap; //顶点映射下标
vector<vector<W>> _matrix; //邻接矩阵
};
}
cpp
void TestGraph1()
{
Graph<char, int> g("0123", 4);
//Graph<char, int, true> g("0123", 4);
g.AddEdge('0', '1', 1);
g.AddEdge('0', '3', 4);
g.AddEdge('1', '3', 2);
g.AddEdge('1', '2', 9);
g.AddEdge('2', '3', 8);
g.AddEdge('2', '1', 5);
g.AddEdge('2', '0', 3);
g.AddEdge('3', '2', 6);
g.Print();
}
邻接矩阵总结:
- 邻接矩阵存储方式非常适合稠密图
- 邻接矩阵O(1)判断两个顶点的连接关系并取到权值
- 相对而言不适合查找一个顶点连接所有边----O(N)
2.邻接表
邻接表:使用数组表示顶点的集合,使用链表表示边的关系
1.无向图邻接表存储
注意:无向图中同一条边在邻接表中出现了两次。如果想知道顶点vi的度,只需要知道顶点vi边链表集合中结点的数目即可
2.有向图邻接表存储
注意:有向图中每条边在邻接表中只出现一次,与顶点vi对应的邻接表所含结点的个数,就是该顶点的出度,也称出度表,要得到vi顶点的入度,必须检测其他所有顶点对应的边链表,看有多少边顶点的dst取值是i
cpp
namespace link_table
{
template<class W>
struct Edge
{
//int _srci;
int _dsti; //目标点的下标
W _w; //权值
Edge<W>* _next;
Edge(int dsti, const W& w) :_dsti(dsti), _w(w), _next(nullptr)
{ }
};
template<class V, class W, bool Direction = false>
class Graph
{
typedef Edge<W> Edge;
public:
Graph(const V* a, size_t n)
{
_vertexs.reserve(n);
for (size_t i = 0; i < n; ++i)
{
_vertexs.push_back(a[i]);
_indexMap[a[i]] = i;
}
_tables.resize(n, nullptr);
}
size_t GetVertexIndex(const V& v)
{
auto it = _indexMap.find(v);
if (it != _indexMap.end())
{
return it->second;
}
else
{
//assert(false);
throw invalid_argument("顶点不存在");
return -1;
}
}
void AddEdge(const V& src, const V& dst, const W& w)
{
size_t srci = GetVertexIndex(src);
size_t dsti = GetVertexIndex(dst);
Edge* eg = new Edge(dsti, w);
eg->_next = _tables[srci];
_tables[srci] = eg;
if (Direction == false)
{
Edge* eg = new Edge(srci, w);
eg->_next = _tables[dsti];
_tables[dsti] = eg;
}
}
void Print()
{
//顶点
for (size_t i = 0; i < _vertexs.size(); ++i)
{
cout << "[" << i << "]" << "->" << _vertexs[i] << endl;
}
cout << endl;
for (size_t i = 0; i < _tables.size(); ++i)
{
cout << _vertexs[i] << "[" << i << "]->";
Edge* cur = _tables[i];
while (cur)
{
cout << "[" << _vertexs[cur->_dsti] << ":" << cur->_dsti << ":" << cur->_w << "]->";
cur = cur->_next;
}
cout << "nullptr" << endl;
}
}
private:
vector<V> _vertexs; //顶点集合
map<V, int> _indexMap; //顶点映射下标
vector<Edge*> _tables; //邻接表
};
}
cpp
void TestGraph2()
{
string a[] = { "张三", "李四", "王五", "赵六" };
//Graph<string, int, true> g1(a, 4);
Graph<string, int> g1(a, 4);
g1.AddEdge("张三", "李四", 100);
g1.AddEdge("张三", "王五", 200);
g1.AddEdge("王五", "赵六", 30);
g1.Print();
}
邻接表总结:
- 适合存储稀疏图
- 适合查找一个顶点连接出去的边
- 不适合确定两个顶点是否相连及权值
三.图的遍历
1.图的广度优先遍历
cpp
void BFS(const V& src)
{
size_t srci = GetVertexIndex(src);
//队列和标记数组
queue<int> q;
vector<bool> visited(_vertexs.size(), false);
q.push(srci);
visited[srci] = true;
int levelSize = 1;
size_t n = _vertexs.size();
while (!q.empty())
{
//一层一层出
for (int i = 0; i < levelSize; ++i)
{
int front = q.front();
q.pop();
cout << front << ":" << _vertexs[front] << " ";
//把front顶点的邻接顶点入队列
for (size_t i = 0; i < n; ++i)
{
if (_matrix[front][i] != MAX_W)
{
if (visited[i] == false)
{
q.push(i);
visited[i] = true;
}
}
}
}
cout << endl;
levelSize = q.size();
}
cout << endl;
}
cpp
void TestBDFS()
{
string a[] = { "张三", "李四", "王五", "赵六", "周七" };
Graph<string, int> g1(a, sizeof(a) / sizeof(string));
g1.AddEdge("张三", "李四", 100);
g1.AddEdge("张三", "王五", 200);
g1.AddEdge("王五", "赵六", 30);
g1.AddEdge("王五", "周七", 30);
g1.Print();
g1.BFS("张三");
}
2.图的深度优先遍历
cpp
void _DFS(size_t srci, vector<bool>& visited)
{
cout << srci << ":" << _vertexs[srci] << endl;
visited[srci] = true;
//找一个和srci相邻的没有访问过的点,去深度遍历
for (size_t i = 0; i < _vertexs.size(); ++i)
{
if (_matrix[srci][i] != MAX_W && visited[i] == false)
{
_DFS(i, visited);
}
}
}
void DFS(const V& src)
{
size_t srci = GetVertexIndex(src);
vector<bool> visited(_vertexs.size(), false);
_DFS(srci, visited);
}
cpp
void TestBDFS()
{
string a[] = { "张三", "李四", "王五", "赵六", "周七" };
Graph<string, int> g1(a, sizeof(a) / sizeof(string));
g1.AddEdge("张三", "李四", 100);
g1.AddEdge("张三", "王五", 200);
g1.AddEdge("王五", "赵六", 30);
g1.AddEdge("王五", "周七", 30);
g1.Print();
g1.DFS("张三");
}
四.最小生成树
连通图中的每一棵生成树,都是原图的一个极大无环子图,即:从其中删去任何一条边,生成树就不在连通;反之,在其中引入任何一条新边,都会形成一条回路
若连通图由n个顶点组成,则其生成树必含n个顶点和n-1条边。因此构造最小生成树的准则有三条:
- 只能使用图中的边来构造最小生成树
- 只能使用恰好n-1条边来连接图中的n个顶点
- 选用的n-1条边不能构成回路
1.Kruskal算法
Kruskal算法(克鲁斯卡尔算法)任给一个有n个顶点的连通网络N={V,E} , 首先构造一个由这n个顶点组成、不含任何边的图G={V,NULL} ,其中每个顶点自成一个连通分量, 其次不断从E中取出权值最小的一条边(若有多条任取其一),若该边的两个顶点来自不同的连通分量,则将此边加入到G中。如此重复,直到所有顶点在同一个连通分量上为止
核心:每次迭代时,选出一条具有最小权值,且两端点不在同一连通分量上的边,加入生成树
cpp
struct Edge
{
size_t _srci;
size_t _dsti;
W _w;
Edge(size_t srci, size_t dsti, const W& w) :_srci(srci), _dsti(dsti), _w(w)
{ }
bool operator>(const Edge& e) const
{
return _w > e._w;
}
};
W Kruskal(Self& minTree)
{
size_t n = _vertexs.size();
minTree._vertexs = _vertexs;
minTree._indexMap = _indexMap;
minTree._matrix.resize(n);
for (size_t i = 0; i < n; ++i)
{
minTree._matrix[i].resize(n, MAX_W);
}
priority_queue<Edge, vector<Edge>, greater<Edge>> minque;
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (i < j && _matrix[i][j] != MAX_W)
{
minque.push(Edge(i, j, _matrix[i][j]));
}
}
}
cout << "Kruskal开始选边:" << endl;
//选出n-1条边
int size = 0;
W totalW = W();
UnionFindSet ufs(n);
while (!minque.empty())
{
Edge min = minque.top();
minque.pop();
if (!ufs.InSet(min._srci, min._dsti))
{
cout << _vertexs[min._srci] << "->" << _vertexs[min._dsti] << ":" << min._w << endl;
minTree._AddEdge(min._srci, min._dsti, min._w);
ufs.Union(min._srci, min._dsti);
++size;
totalW += min._w;
}
else
{
cout << "构成环";
cout << _vertexs[min._srci] << "->" << _vertexs[min._dsti] << ":" << min._w << endl;
}
}
cout << endl;
if (size == n - 1)
{
return totalW;
}
else
{
return W();
}
}
cpp
void TestGraphMinTree()
{
const char str[] = "abcdefghi";
Graph<char, int> g(str, strlen(str));
g.AddEdge('a', 'b', 4);
g.AddEdge('a', 'h', 8);
g.AddEdge('a', 'h', 9);
g.AddEdge('b', 'c', 8);
g.AddEdge('b', 'h', 11);
g.AddEdge('c', 'i', 2);
g.AddEdge('c', 'f', 4);
g.AddEdge('c', 'd', 7);
g.AddEdge('d', 'f', 14);
g.AddEdge('d', 'e', 9);
g.AddEdge('e', 'f', 10);
g.AddEdge('f', 'g', 2);
g.AddEdge('g', 'h', 1);
g.AddEdge('g', 'i', 6);
g.AddEdge('h', 'i', 7);
Graph<char, int> kminTree;
cout << "Kruskal:" << g.Kruskal(kminTree) << endl;
kminTree.Print();
cout << endl;
}
2.Prim算法
Prim算法(普里姆算法)
cpp
W Prim(Self& minTree, const W& src)
{
size_t srci = GetVertexIndex(src);
size_t n = _vertexs.size();
minTree._vertexs = _vertexs;
minTree._indexMap = _indexMap;
minTree._matrix.resize(n);
for (size_t i = 0; i < n; ++i)
{
minTree._matrix[i].resize(n, MAX_W);
}
vector<bool> X(n, false);
vector<bool> Y(n, true);
X[srci] = true;
Y[srci] = false;
//从X到Y集合相连接的边中选出值最小的边
priority_queue<Edge, vector<Edge>, greater<Edge>> minq;
//将srci连接的边添加到队列中
for (size_t i = 0; i < n; ++i)
{
if (_matrix[srci][i] != MAX_W)
{
minq.push(Edge(srci, i, _matrix[srci][i]));
}
}
cout << "Prim开始选边:" << endl;
int size = 0;
W totalW = W();
while (!minq.empty())
{
Edge min = minq.top();
minq.pop();
if (X[min._dsti])
{
cout << "构成环";
cout << _vertexs[min._srci] << "->" << _vertexs[min._dsti] << ":" << min._w << endl;
}
else
{
minTree._AddEdge(min._srci, min._dsti, min._w);
cout << _vertexs[min._srci] << "->" << _vertexs[min._dsti] << ":" << min._w << endl;
X[min._dsti] = true;
Y[min._dsti] = false;
++size;
totalW += min._w;
if (size == n - 1)
break;
for (size_t i = 0; i < n; ++i)
{
if (_matrix[min._dsti][i] != MAX_W && Y[i])
{
minq.push(Edge(min._dsti, i, _matrix[min._dsti][i]));
}
}
}
}
cout << endl;
if (size == n - 1)
{
return totalW;
}
else
{
return W();
}
}
cpp
void TestGraphMinTree()
{
const char str[] = "abcdefghi";
Graph<char, int> g(str, strlen(str));
g.AddEdge('a', 'b', 4);
g.AddEdge('a', 'h', 8);
g.AddEdge('a', 'h', 9);
g.AddEdge('b', 'c', 8);
g.AddEdge('b', 'h', 11);
g.AddEdge('c', 'i', 2);
g.AddEdge('c', 'f', 4);
g.AddEdge('c', 'd', 7);
g.AddEdge('d', 'f', 14);
g.AddEdge('d', 'e', 9);
g.AddEdge('e', 'f', 10);
g.AddEdge('f', 'g', 2);
g.AddEdge('g', 'h', 1);
g.AddEdge('g', 'i', 6);
g.AddEdge('h', 'i', 7);
Graph<char, int> pminTree;
cout << "Prim:" << g.Prim(pminTree, 'a') << endl;
pminTree.Print();
}
五.最短路径
最短路径问题:从在带权有向图G中的某一顶点出发,找出一条通往另一顶点的最短路径,最短也就是沿路径各边的权值总和达到最小
1.单源最短路径--Dijkstra算法
单源最短路径问题:给定一个图G = ( V , E ) G=(V,E)G=(V,E),求源结点s ∈ V s∈Vs∈V到图中每个结点v ∈ V v∈Vv∈V的最短路径。Dijkstra算法就适用于解决带权重的有向图上的单源最短路径问题,同时算法要求图中所有边的权重非负。一般在求解最短路径的时候都是已知一个起点和一个终点,所以使用Dijkstra算法求解过后也就得到了所需起点到终点的最短路径
针对一个带权有向图G,将所有结点分为两组S和Q,S是已经确定最短路径的结点集合,在初始时为空(初始时就可以将源节点s放入,毕竟源节点到自己的代价是0),Q 为其余未确定最短路径 的结点集合,每次从Q 中找出一个起点到该结点代价最小的结点u ,将u 从Q 中移出,并放入S中,对u的每一个相邻结点v 进行松弛操作。松弛即对每一个相邻结点v ,判断源节点s到结点u 的代价与u 到v 的代价之和是否比原来s 到v 的代价更小,若代价比原来小则要将s 到v 的代价更新 为s 到u 与u 到v 的代价之和,否则维持原样。如此一直循环直至集合Q 为空,即所有节点都已经 查找过一遍并确定了最短路径,至于一些起点到达不了的结点在算法循环后其代价仍为初始设定 的值,不发生变化。Dijkstra算法每次都是选择V-S中最小的路径节点来进行更新,并加入S中,所 以该算法使用的是贪心策略
Dijkstra算法(迪杰斯特拉算法)存在的问题是不支持图中带负权路径,如果带有负权路径,则可能会找不到一些路径的最短路径
cpp
//时间复杂度:O(N^2),空间复杂度:O(N)
void Dijkstra(const V& src, vector<W>& dist, vector<int>& pPath)
{
size_t srci = GetVertexIndex(src);
size_t n = _vertexs.size();
dist.resize(n, MAX_W);
pPath.resize(n, -1);
dist[srci] = 0;
pPath[srci] = srci;
//已经确定最短路径的顶点集合
vector<bool> S(n, false);
for (size_t j = 0; j < n; ++j)
{
//选出最短路径顶点且不在S更新其他路径
int u = 0;
W min = MAX_W;
for (size_t i = 0; i < n; ++i)
{
if (S[i] == false && dist[i] < min)
{
u = i;
min = dist[i];
}
}
S[u] = true;
//松弛更新
for (size_t v = 0; v < n; ++v)
{
if (S[v] == false && _matrix[u][v] != MAX_W && dist[u] + _matrix[u][v] < dist[v])
{
dist[v] = dist[u] + _matrix[u][v];
pPath[v] = u;
}
}
}
}
cpp
void TestGraphDijkstra()
{
const char* str = "syztx";
Graph<char, int, INT_MAX, true> g(str, strlen(str));
g.AddEdge('s', 't', 10);
g.AddEdge('s', 'y', 5);
g.AddEdge('y', 't', 3);
g.AddEdge('y', 'x', 9);
g.AddEdge('y', 'z', 2);
g.AddEdge('z', 's', 7);
g.AddEdge('z', 'x', 6);
g.AddEdge('t', 'y', 2);
g.AddEdge('t', 'x', 1);
g.AddEdge('x', 'z', 4);
vector<int> dist;
vector<int> parentPath;
g.Dijkstra('s', dist, parentPath);
g.PrintShortPath('s', dist, parentPath);
}
2.单源最短路径--Bellman-Ford算法
Dijkstra算法只能用来解决正权图的单源最短路径问题,但有些题目会出现负权图。这时这个算法就不能帮助我们解决问题了,而bellman---ford算法(贝尔曼-福特算法)可以解决负权图的单源最短路径问题。它的优点是可以解决有负权边的单源最短路径问题,而且可以用来判断是否有负权回路 。它也有明显的缺点,它的时间复杂度 O(N*E) (N是点数,E是边数)普遍是要高于Dijkstra算法O(N²)的。**像这里 如果我们使用邻接矩阵实现,那么遍历所有边的数量的时间复杂度就是O(N^3),**这里也可以看出来Bellman-Ford就是一种暴力求解更新
cpp
//时间复杂度:O(N^3),空间复杂度:O(N)
bool BellmanFord(const V& src, vector<W>& dist, vector<int>& pPath)
{
size_t srci = GetVertexIndex(src);
size_t n = _vertexs.size();
// vector<W> dist,记录srci-其他顶点最短路径权值数组
dist.resize(n, MAX_W);
// vector<int> pPath 记录srci-其他顶点最短路径父顶点数组
pPath.resize(n, -1);
// 先更新srci->srci为最小值
dist[srci] = W();
for (size_t k = 0; k < n; ++k)
{
bool updata = false;
cout << "更新第" << k << "轮:" << endl;
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (_matrix[i][j] != MAX_W && dist[i] + _matrix[i][j] < dist[j])
{
updata = true;
cout << _vertexs[i] << "->" << _vertexs[j] << ":" << _matrix[i][j] << endl;
dist[j] = dist[i] + _matrix[i][j];
pPath[j] = i;
}
}
}
//如果这个轮次没有更新出最短路径,后续轮次就不需要再走
if (updata == false)
break;
}
//如果还能更新就是带负权回路
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (_matrix[i][j] != MAX_W && dist[i] + _matrix[i][j] < dist[j])
{
return false;
}
}
}
return true;
}
cpp
void TestGraphBellmanFord()
{
const char* str = "syztx";
Graph<char, int, INT_MAX, true> g(str, strlen(str));
g.AddEdge('s', 't', 6);
g.AddEdge('s', 'y', 7);
g.AddEdge('y', 'z', 9);
g.AddEdge('y', 'x', -3);
//g.AddEdge('y', 's', 1);//新增
g.AddEdge('z', 's', 2);
g.AddEdge('z', 'x', 7);
g.AddEdge('t', 'x', 5);
g.AddEdge('t', 'y', 8);
//g.AddEdge('t', 'y', -8);//更改
g.AddEdge('t', 'z', -4);
g.AddEdge('x', 't', -2);
vector<int> dist;
vector<int> parentPath;
if (g.BellmanFord('s', dist, parentPath))
{
g.PrintShortPath('s', dist, parentPath);
}
else
{
cout << "存在负权回路" << endl;
}
}
3.多源最短路径--Floyd-Warshall算法
Floyd-Warshall算法(弗洛伊德算法)是解决任意两点间的最短路径的一种算法
Floyd算法考虑的是一条最短路径的中间节点,即简单路径p={v1,v2,...,vn}上除v1和vn的任意节点
设k是p的一个中间节点,那么从i到j的最短路径p就被分成i到k和k到j的两段最短路径p1,p2。p1 是从i到k且中间节点属于{1,2,...,k-1}取得的一条最短路径。p2是从k到j且中间节点属于{1, 2,...,k-1}取得的一条最短路径
Floyd算法本质是三维动态规划,D[i][j][k]表示从点i到点j只经过0到k个点最短路径,然后建立 起转移方程,然后通过空间优化,优化掉最后一维度,变成一个最短路径的迭代算法,最后即得到所有点的最短路径
cpp
void FloydWarShall(vector<vector<W>>& vvDist, vector<vector<int>>& vvpPath)
{
size_t n = _vertexs.size();
vvDist.resize(n);
vvpPath.resize(n);
for (size_t i = 0; i < n; ++i)
{
vvDist[i].resize(n, MAX_W);
vvpPath[i].resize(n, -1);
}
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (_matrix[i][j] != MAX_W)
{
vvDist[i][j] = _matrix[i][j];
vvpPath[i][j] = i;
}
if (i == j)
{
vvDist[i][j] = 0;
}
}
}
//最短路径的更新
for (size_t k = 0; k < n; ++k)
{
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (vvDist[i][k] != MAX_W && vvDist[k][j] != MAX_W && vvDist[i][k] + vvDist[k][j] < vvDist[i][j])
{
vvDist[i][j] = vvDist[i][k] + vvDist[k][j];
vvpPath[i][j] = vvpPath[k][j];
}
}
}
}
}
cpp
void TestFloydWarShall()
{
const char* str = "12345";
Graph<char, int, INT_MAX, true> g(str, strlen(str));
g.AddEdge('1', '2', 3);
g.AddEdge('1', '3', 8);
g.AddEdge('1', '5', -4);
g.AddEdge('2', '4', 1);
g.AddEdge('2', '5', 7);
g.AddEdge('3', '2', 4);
g.AddEdge('4', '1', 2);
g.AddEdge('4', '3', -5);
g.AddEdge('5', '4', 6);
vector<vector<int>> vvDist;
vector<vector<int>> vvParentPath;
g.FloydWarShall(vvDist, vvParentPath);
// 打印任意两点之间的最短路径
for (size_t i = 0; i < strlen(str); ++i)
{
g.PrintShortPath(str[i], vvDist[i], vvParentPath[i]);
cout << endl;
}
}
六.整体实现
1.UnionFindSet.h
cpp
#pragma once
#include<vector>
class UnionFindSet
{
public:
UnionFindSet(size_t n) :_ufs(n, -1)
{ }
int FindRoot(int x)
{
int root = x;
while (_ufs[root] >= 0)
root = _ufs[root];
//路径压缩
while (_ufs[x] >= 0)
{
int parent = _ufs[x];
_ufs[x] = root;
x = parent;
}
return root;
}
bool Union(int x1, int x2)
{
int root1 = FindRoot(x1);
int root2 = FindRoot(x2);
if (root1 == root2)//x1和x2本来就在一个集合中
return false;
//数据量小的向大的合并
if (abs(_ufs[root1]) < abs(_ufs[root2]))
swap(root1, root2);
_ufs[root1] += _ufs[root2];
_ufs[root2] = root1;
return true;
}
bool InSet(int x1, int x2)
{
return FindRoot(x1) == FindRoot(x2);
}
size_t SetSize()
{
size_t n = 0;
for (auto& e : _ufs)
{
if (e < 0)
++n;
}
return n;
}
private:
vector<int> _ufs;
};
void TestUoionFindSet()
{
UnionFindSet ufs(10);
ufs.Union(8, 9);
ufs.Union(7, 8);
ufs.Union(6, 7);
ufs.Union(5, 6);
ufs.Union(4, 5);
}
2.Graph.h
cpp
#pragma once
#include<vector>
#include<string>
#include<map>
#include<queue>
#include<set>
#include<functional>
namespace matrix
{
template<class V, class W, W MAX_W = INT_MAX, bool Direction = false>
class Graph
{
typedef Graph<V, W, MAX_W, Direction> Self;
public:
//图的创建
//1.IO输入->不方便测试,oj中更适合
//2.图结构关系写到文件,读取文件
//3.手动添加边
Graph() = default;
Graph(const V* a, size_t n)
{
_vertexs.reserve(n);
for (size_t i = 0; i < n; ++i)
{
_vertexs.push_back(a[i]);
_indexMap[a[i]] = i;
}
_matrix.resize(n);
for (size_t i = 0; i < _matrix.size(); ++i)
{
_matrix[i].resize(n, MAX_W);
}
}
size_t GetVertexIndex(const V& v)
{
auto it = _indexMap.find(v);
if (it != _indexMap.end())
{
return it->second;
}
else
{
//assert(false);
throw invalid_argument("顶点不存在");
return -1;
}
}
void _AddEdge(size_t srci, size_t dsti, const W& w)
{
_matrix[srci][dsti] = w;
//无向图
if (Direction == false)
{
_matrix[dsti][srci] = w;
}
}
void AddEdge(const V& src, const V& dst, const W& w)
{
size_t srci = GetVertexIndex(src);
size_t dsti = GetVertexIndex(dst);
_AddEdge(srci, dsti, w);
}
void Print()
{
//顶点
for (size_t i = 0; i < _vertexs.size(); ++i)
{
cout << "[" << i << "]" << "->" << _vertexs[i] << endl;
}
cout << endl;
//矩阵
//横下标
cout << " ";
for (size_t i = 0; i < _matrix.size(); ++i)
{
//cout << i << " ";
printf("%4d", i);
}
cout << endl;
for (size_t i = 0; i < _matrix.size(); ++i)
{
cout << i << " ";//竖下标
for (size_t j = 0; j < _matrix[i].size(); ++j)
{
//cout << _matrix[i][j] << " ";
if (_matrix[i][j] == MAX_W)
{
//cout << "* ";
printf("%4c", '*');
}
else
{
//cout << _matrix[i][j] << " ";
printf("%4d", _matrix[i][j]);
}
}
cout << endl;
}
cout << endl;
}
void BFS(const V& src)
{
size_t srci = GetVertexIndex(src);
//队列和标记数组
queue<int> q;
vector<bool> visited(_vertexs.size(), false);
q.push(srci);
visited[srci] = true;
int levelSize = 1;
size_t n = _vertexs.size();
while (!q.empty())
{
//一层一层出
for (int i = 0; i < levelSize; ++i)
{
int front = q.front();
q.pop();
cout << front << ":" << _vertexs[front] << " ";
//把front顶点的邻接顶点入队列
for (size_t i = 0; i < n; ++i)
{
if (_matrix[front][i] != MAX_W)
{
if (visited[i] == false)
{
q.push(i);
visited[i] = true;
}
}
}
}
cout << endl;
levelSize = q.size();
}
cout << endl;
}
void _DFS(size_t srci, vector<bool>& visited)
{
cout << srci << ":" << _vertexs[srci] << endl;
visited[srci] = true;
//找一个和srci相邻的没有访问过的点,去深度遍历
for (size_t i = 0; i < _vertexs.size(); ++i)
{
if (_matrix[srci][i] != MAX_W && visited[i] == false)
{
_DFS(i, visited);
}
}
}
void DFS(const V& src)
{
size_t srci = GetVertexIndex(src);
vector<bool> visited(_vertexs.size(), false);
_DFS(srci, visited);
}
struct Edge
{
size_t _srci;
size_t _dsti;
W _w;
Edge(size_t srci, size_t dsti, const W& w) :_srci(srci), _dsti(dsti), _w(w)
{ }
bool operator>(const Edge& e) const
{
return _w > e._w;
}
};
W Kruskal(Self& minTree)
{
size_t n = _vertexs.size();
minTree._vertexs = _vertexs;
minTree._indexMap = _indexMap;
minTree._matrix.resize(n);
for (size_t i = 0; i < n; ++i)
{
minTree._matrix[i].resize(n, MAX_W);
}
priority_queue<Edge, vector<Edge>, greater<Edge>> minque;
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (i < j && _matrix[i][j] != MAX_W)
{
minque.push(Edge(i, j, _matrix[i][j]));
}
}
}
cout << "Kruskal开始选边:" << endl;
//选出n-1条边
int size = 0;
W totalW = W();
UnionFindSet ufs(n);
while (!minque.empty())
{
Edge min = minque.top();
minque.pop();
if (!ufs.InSet(min._srci, min._dsti))
{
cout << _vertexs[min._srci] << "->" << _vertexs[min._dsti] << ":" << min._w << endl;
minTree._AddEdge(min._srci, min._dsti, min._w);
ufs.Union(min._srci, min._dsti);
++size;
totalW += min._w;
}
else
{
cout << "构成环";
cout << _vertexs[min._srci] << "->" << _vertexs[min._dsti] << ":" << min._w << endl;
}
}
cout << endl;
if (size == n - 1)
{
return totalW;
}
else
{
return W();
}
}
W Prim(Self& minTree, const W& src)
{
size_t srci = GetVertexIndex(src);
size_t n = _vertexs.size();
minTree._vertexs = _vertexs;
minTree._indexMap = _indexMap;
minTree._matrix.resize(n);
for (size_t i = 0; i < n; ++i)
{
minTree._matrix[i].resize(n, MAX_W);
}
vector<bool> X(n, false);
vector<bool> Y(n, true);
X[srci] = true;
Y[srci] = false;
//从X到Y集合相连接的边中选出值最小的边
priority_queue<Edge, vector<Edge>, greater<Edge>> minq;
//将srci连接的边添加到队列中
for (size_t i = 0; i < n; ++i)
{
if (_matrix[srci][i] != MAX_W)
{
minq.push(Edge(srci, i, _matrix[srci][i]));
}
}
cout << "Prim开始选边:" << endl;
int size = 0;
W totalW = W();
while (!minq.empty())
{
Edge min = minq.top();
minq.pop();
if (X[min._dsti])
{
cout << "构成环";
cout << _vertexs[min._srci] << "->" << _vertexs[min._dsti] << ":" << min._w << endl;
}
else
{
minTree._AddEdge(min._srci, min._dsti, min._w);
cout << _vertexs[min._srci] << "->" << _vertexs[min._dsti] << ":" << min._w << endl;
X[min._dsti] = true;
Y[min._dsti] = false;
++size;
totalW += min._w;
if (size == n - 1)
break;
for (size_t i = 0; i < n; ++i)
{
if (_matrix[min._dsti][i] != MAX_W && Y[i])
{
minq.push(Edge(min._dsti, i, _matrix[min._dsti][i]));
}
}
}
}
cout << endl;
if (size == n - 1)
{
return totalW;
}
else
{
return W();
}
}
void PrintShortPath(const V& src, const vector<W>& dist, const vector<int>& pPath)
{
size_t srci = GetVertexIndex(src);
size_t n = _vertexs.size();
for (size_t i = 0; i < n; ++i)
{
if (i != srci)
{
//找出i顶点的路径
vector<int> path;
size_t parenti = i;
while (parenti != srci)
{
path.push_back(parenti);
parenti = pPath[parenti];
}
path.push_back(srci);
reverse(path.begin(), path.end());
for (auto index : path)
{
cout << _vertexs[index] << "->";
}
cout << "权值和: " << dist[i] << endl;
}
}
}
//时间复杂度:O(N^2),空间复杂度:O(N)
void Dijkstra(const V& src, vector<W>& dist, vector<int>& pPath)
{
size_t srci = GetVertexIndex(src);
size_t n = _vertexs.size();
dist.resize(n, MAX_W);
pPath.resize(n, -1);
dist[srci] = 0;
pPath[srci] = srci;
//已经确定最短路径的顶点集合
vector<bool> S(n, false);
for (size_t j = 0; j < n; ++j)
{
//选出最短路径顶点且不在S更新其他路径
int u = 0;
W min = MAX_W;
for (size_t i = 0; i < n; ++i)
{
if (S[i] == false && dist[i] < min)
{
u = i;
min = dist[i];
}
}
S[u] = true;
//松弛更新
for (size_t v = 0; v < n; ++v)
{
if (S[v] == false && _matrix[u][v] != MAX_W && dist[u] + _matrix[u][v] < dist[v])
{
dist[v] = dist[u] + _matrix[u][v];
pPath[v] = u;
}
}
}
}
//时间复杂度:O(N^3),空间复杂度:O(N)
bool BellmanFord(const V& src, vector<W>& dist, vector<int>& pPath)
{
size_t srci = GetVertexIndex(src);
size_t n = _vertexs.size();
// vector<W> dist,记录srci-其他顶点最短路径权值数组
dist.resize(n, MAX_W);
// vector<int> pPath 记录srci-其他顶点最短路径父顶点数组
pPath.resize(n, -1);
// 先更新srci->srci为最小值
dist[srci] = W();
for (size_t k = 0; k < n; ++k)
{
bool updata = false;
cout << "更新第" << k << "轮:" << endl;
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (_matrix[i][j] != MAX_W && dist[i] + _matrix[i][j] < dist[j])
{
updata = true;
cout << _vertexs[i] << "->" << _vertexs[j] << ":" << _matrix[i][j] << endl;
dist[j] = dist[i] + _matrix[i][j];
pPath[j] = i;
}
}
}
//如果这个轮次没有更新出最短路径,后续轮次就不需要再走
if (updata == false)
break;
}
//如果还能更新就是带负权回路
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (_matrix[i][j] != MAX_W && dist[i] + _matrix[i][j] < dist[j])
{
return false;
}
}
}
return true;
}
void FloydWarShall(vector<vector<W>>& vvDist, vector<vector<int>>& vvpPath)
{
size_t n = _vertexs.size();
vvDist.resize(n);
vvpPath.resize(n);
for (size_t i = 0; i < n; ++i)
{
vvDist[i].resize(n, MAX_W);
vvpPath[i].resize(n, -1);
}
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (_matrix[i][j] != MAX_W)
{
vvDist[i][j] = _matrix[i][j];
vvpPath[i][j] = i;
}
if (i == j)
{
vvDist[i][j] = 0;
}
}
}
//最短路径的更新
for (size_t k = 0; k < n; ++k)
{
for (size_t i = 0; i < n; ++i)
{
for (size_t j = 0; j < n; ++j)
{
if (vvDist[i][k] != MAX_W && vvDist[k][j] != MAX_W && vvDist[i][k] + vvDist[k][j] < vvDist[i][j])
{
vvDist[i][j] = vvDist[i][k] + vvDist[k][j];
vvpPath[i][j] = vvpPath[k][j];
}
}
}
}
}
private:
vector<V> _vertexs; //顶点集合
map<V, int> _indexMap; //顶点映射下标
vector<vector<W>> _matrix; //邻接矩阵
};
void TestGraph1()
{
Graph<char, int> g("0123", 4);
//Graph<char, int, true> g("0123", 4);
g.AddEdge('0', '1', 1);
g.AddEdge('0', '3', 4);
g.AddEdge('1', '3', 2);
g.AddEdge('1', '2', 9);
g.AddEdge('2', '3', 8);
g.AddEdge('2', '1', 5);
g.AddEdge('2', '0', 3);
g.AddEdge('3', '2', 6);
g.Print();
}
void TestBDFS()
{
string a[] = { "张三", "李四", "王五", "赵六", "周七" };
Graph<string, int> g1(a, sizeof(a) / sizeof(string));
g1.AddEdge("张三", "李四", 100);
g1.AddEdge("张三", "王五", 200);
g1.AddEdge("王五", "赵六", 30);
g1.AddEdge("王五", "周七", 30);
g1.Print();
g1.BFS("张三");
g1.DFS("张三");
}
void TestGraphMinTree()
{
const char str[] = "abcdefghi";
Graph<char, int> g(str, strlen(str));
g.AddEdge('a', 'b', 4);
g.AddEdge('a', 'h', 8);
g.AddEdge('a', 'h', 9);
g.AddEdge('b', 'c', 8);
g.AddEdge('b', 'h', 11);
g.AddEdge('c', 'i', 2);
g.AddEdge('c', 'f', 4);
g.AddEdge('c', 'd', 7);
g.AddEdge('d', 'f', 14);
g.AddEdge('d', 'e', 9);
g.AddEdge('e', 'f', 10);
g.AddEdge('f', 'g', 2);
g.AddEdge('g', 'h', 1);
g.AddEdge('g', 'i', 6);
g.AddEdge('h', 'i', 7);
Graph<char, int> kminTree;
cout << "Kruskal:" << g.Kruskal(kminTree) << endl;
kminTree.Print();
cout << endl;
Graph<char, int> pminTree;
cout << "Prim:" << g.Prim(pminTree, 'a') << endl;
pminTree.Print();
}
void TestGraphDijkstra()
{
const char* str = "syztx";
Graph<char, int, INT_MAX, true> g(str, strlen(str));
g.AddEdge('s', 't', 10);
g.AddEdge('s', 'y', 5);
g.AddEdge('y', 't', 3);
g.AddEdge('y', 'x', 9);
g.AddEdge('y', 'z', 2);
g.AddEdge('z', 's', 7);
g.AddEdge('z', 'x', 6);
g.AddEdge('t', 'y', 2);
g.AddEdge('t', 'x', 1);
g.AddEdge('x', 'z', 4);
vector<int> dist;
vector<int> parentPath;
g.Dijkstra('s', dist, parentPath);
g.PrintShortPath('s', dist, parentPath);
}
void TestGraphBellmanFord()
{
const char* str = "syztx";
Graph<char, int, INT_MAX, true> g(str, strlen(str));
g.AddEdge('s', 't', 6);
g.AddEdge('s', 'y', 7);
g.AddEdge('y', 'z', 9);
g.AddEdge('y', 'x', -3);
//g.AddEdge('y', 's', 1);//新增
g.AddEdge('z', 's', 2);
g.AddEdge('z', 'x', 7);
g.AddEdge('t', 'x', 5);
g.AddEdge('t', 'y', 8);
//g.AddEdge('t', 'y', -8);//更改
g.AddEdge('t', 'z', -4);
g.AddEdge('x', 't', -2);
vector<int> dist;
vector<int> parentPath;
if (g.BellmanFord('s', dist, parentPath))
{
g.PrintShortPath('s', dist, parentPath);
}
else
{
cout << "存在负权回路" << endl;
}
}
void TestFloydWarShall()
{
const char* str = "12345";
Graph<char, int, INT_MAX, true> g(str, strlen(str));
g.AddEdge('1', '2', 3);
g.AddEdge('1', '3', 8);
g.AddEdge('1', '5', -4);
g.AddEdge('2', '4', 1);
g.AddEdge('2', '5', 7);
g.AddEdge('3', '2', 4);
g.AddEdge('4', '1', 2);
g.AddEdge('4', '3', -5);
g.AddEdge('5', '4', 6);
vector<vector<int>> vvDist;
vector<vector<int>> vvParentPath;
g.FloydWarShall(vvDist, vvParentPath);
// 打印任意两点之间的最短路径
for (size_t i = 0; i < strlen(str); ++i)
{
g.PrintShortPath(str[i], vvDist[i], vvParentPath[i]);
cout << endl;
}
}
}
namespace link_table
{
template<class W>
struct Edge
{
//int _srci;
int _dsti; //目标点的下标
W _w; //权值
Edge<W>* _next;
Edge(int dsti, const W& w) :_dsti(dsti), _w(w), _next(nullptr)
{ }
};
template<class V, class W, bool Direction = false>
class Graph
{
typedef Edge<W> Edge;
public:
Graph(const V* a, size_t n)
{
_vertexs.reserve(n);
for (size_t i = 0; i < n; ++i)
{
_vertexs.push_back(a[i]);
_indexMap[a[i]] = i;
}
_tables.resize(n, nullptr);
}
size_t GetVertexIndex(const V& v)
{
auto it = _indexMap.find(v);
if (it != _indexMap.end())
{
return it->second;
}
else
{
//assert(false);
throw invalid_argument("顶点不存在");
return -1;
}
}
void AddEdge(const V& src, const V& dst, const W& w)
{
size_t srci = GetVertexIndex(src);
size_t dsti = GetVertexIndex(dst);
Edge* eg = new Edge(dsti, w);
eg->_next = _tables[srci];
_tables[srci] = eg;
if (Direction == false)
{
Edge* eg = new Edge(srci, w);
eg->_next = _tables[dsti];
_tables[dsti] = eg;
}
}
void Print()
{
//顶点
for (size_t i = 0; i < _vertexs.size(); ++i)
{
cout << "[" << i << "]" << "->" << _vertexs[i] << endl;
}
cout << endl;
for (size_t i = 0; i < _tables.size(); ++i)
{
cout << _vertexs[i] << "[" << i << "]->";
Edge* cur = _tables[i];
while (cur)
{
cout << "[" << _vertexs[cur->_dsti] << ":" << cur->_dsti << ":" << cur->_w << "]->";
cur = cur->_next;
}
cout << "nullptr" << endl;
}
}
private:
vector<V> _vertexs; //顶点集合
map<V, int> _indexMap; //顶点映射下标
vector<Edge*> _tables; //邻接表
};
void TestGraph2()
{
string a[] = { "张三", "李四", "王五", "赵六" };
//Graph<string, int, true> g1(a, 4);
Graph<string, int> g1(a, 4);
g1.AddEdge("张三", "李四", 100);
g1.AddEdge("张三", "王五", 200);
g1.AddEdge("王五", "赵六", 30);
g1.Print();
}
}
3.test.cpp
cpp
#include<iostream>
using namespace std;
#include"UnionFindSet.h"
#include"Graph.h"
int main()
{
//TestUoionFindSet();
//matrix::TestGraph1();
//matrix::TestBDFS();
//matrix::TestGraphMinTree();
//matrix::TestGraphDijkstra();
//matrix::TestGraphBellmanFord();
matrix::TestFloydWarShall();
//link_table::TestGraph2();
return 0;
}