图论在数据结构中是非常有趣而复杂的,作为web码农的我,在实际开发中一直没有找到它的使用场景,不像树那样的频繁使用,不过还是准备
仔细的把图论全部过一遍。
一:最小生成树
图中有一个好玩的东西叫做生成树,就是用边来把所有的顶点联通起来,前提条件是最后形成的联通图中不能存在回路,所以就形成这样一个
推理:假设图中的顶点有n个,则生成树的边有n-1条,多一条会存在回路,少一路则不能把所有顶点联通起来,如果非要在图中加上权重,则生成树
中权重最小的叫做最小生成树。
对于上面这个带权无向图来说,它的生成树有多个,同样最小生成树也有多个,因为我们比的是权重的大小。
二:Prim算法
求最小生成树的算法有很多,常用的是Prim算法和Kruskal算法,为了保证单一职责,我把Kruskal算法放到下一篇,那么Prim算法的思想
是什么呢?很简单,贪心思想。
如上图:现有集合M={A,B,C,D,E,F},再设集合N={}。
第一步:挑选任意节点(比如A),将其加入到N集合,同时剔除M集合的A。
第二步:寻找A节点权值最小的邻节点(比如F),然后将F加入到N集合,此时N={A,F},同时剔除M集合中的F。
第三步:寻找{A,F}中的权值最小的邻节点(比如E),然后将E加入到N集合,此时N={A,F,E},同时剔除M集合的E。
。。。
最后M集合为{}时,生成树就构建完毕了,是不是非常的简单,这种贪心做法我想大家都能想得到,如果算法配合一个好的数据结构,就会
如虎添翼。
三:代码
1. 图的存储
图的存储有很多方式,邻接矩阵,邻接表,十字链表等等,当然都有自己的适合场景,下面用邻接矩阵来玩玩,邻接矩阵需要采用两个数组,
①. 保存顶点信息的一维数组,
②. 保存边信息的二维数组。
1 public class Graph 2 { 3 /// <summary> 4 /// 顶点个数 5 /// </summary> 6 public char[] vertexs; 7 8 /// <summary> 9 /// 边的条数 10 /// </summary> 11 public int[,] edges; 12 13 /// <summary> 14 /// 顶点个数 15 /// </summary> 16 public int vertexsNum; 17 18 /// <summary> 19 /// 边的个数 20 /// </summary> 21 public int edgesNum; 22 }
2:矩阵构建
矩阵构建很简单,这里把上图中的顶点和权的信息保存在矩阵中。
1 #region 矩阵的构建 2 /// <summary> 3 /// 矩阵的构建 4 /// </summary> 5 public void Build() 6 { 7 //顶点数 8 graph.vertexsNum = 6; 9 10 //边数 11 graph.edgesNum = 8; 12 13 graph.vertexs = new char[graph.vertexsNum]; 14 15 graph.edges = new int[graph.vertexsNum, graph.vertexsNum]; 16 17 //构建二维数组 18 for (int i = 0; i < graph.vertexsNum; i++) 19 { 20 //顶点 21 graph.vertexs[i] = (char)(i + 65); 22 23 for (int j = 0; j < graph.vertexsNum; j++) 24 { 25 graph.edges[i, j] = int.MaxValue; 26 } 27 } 28 29 graph.edges[0, 1] = graph.edges[1, 0] = 80; 30 graph.edges[0, 3] = graph.edges[3, 0] = 100; 31 graph.edges[0, 5] = graph.edges[5, 0] = 20; 32 graph.edges[1, 2] = graph.edges[2, 1] = 90; 33 graph.edges[2, 5] = graph.edges[5, 2] = 70; 34 graph.edges[3, 2] = graph.edges[2, 3] = 100; 35 graph.edges[4, 5] = graph.edges[5, 4] = 40; 36 graph.edges[3, 4] = graph.edges[4, 3] = 60; 37 graph.edges[2, 3] = graph.edges[3, 2] = 10; 38 } 39 #endregion
3:Prim
要玩Prim,我们需要两个字典。
①:保存当前节点的字典,其中包含该节点的起始边和终边以及权值,用weight=-1来记录当前节点已经访问过,用weight=int.MaxValue表示
两节点没有边。
②:输出节点的字典,存放的就是我们的N集合。
当然这个复杂度玩高了,为O(N2),寻找N集合的邻边最小权值时,我们可以玩玩AVL或者优先队列来降低复杂度。
1 #region prim算法 2 /// <summary> 3 /// prim算法 4 /// </summary> 5 public Dictionary<char, Edge> Prim() 6 { 7 Dictionary<char, Edge> dic = new Dictionary<char, Edge>(); 8 9 //统计结果 10 Dictionary<char, Edge> outputDic = new Dictionary<char, Edge>(); 11 12 //weight=MaxValue:标识没有边 13 for (int i = 0; i < graph.vertexsNum; i++) 14 { 15 //起始边 16 var startEdge = (char)(i + 65); 17 18 dic.Add(startEdge, new Edge() { weight = int.MaxValue }); 19 } 20 21 //取字符的开始位置 22 var index = 65; 23 24 //取当前要使用的字符 25 var start = (char)(index); 26 27 for (int i = 0; i < graph.vertexsNum; i++) 28 { 29 //标记开始边已使用过 30 dic[start].weight = -1; 31 32 for (int j = 1; j < graph.vertexsNum; j++) 33 { 34 //获取当前 c 的 邻边 35 var end = (char)(j + index); 36 37 //取当前字符的权重 38 var weight = graph.edges[(int)(start) - index, j]; 39 40 if (weight < dic[end].weight) 41 { 42 dic[end] = new Edge() 43 { 44 weight = weight, 45 startEdge = start, 46 endEdge = end 47 }; 48 } 49 } 50 51 var min = int.MaxValue; 52 53 char minkey = ' '; 54 55 foreach (var key in dic.Keys) 56 { 57 //取当前 最小的 key(使用过的除外) 58 if (min > dic[key].weight && dic[key].weight != -1) 59 { 60 min = dic[key].weight; 61 minkey = key; 62 } 63 } 64 65 start = minkey; 66 67 //边为顶点减去1 68 if (outputDic.Count < graph.vertexsNum - 1 && !outputDic.ContainsKey(minkey)) 69 { 70 outputDic.Add(minkey, new Edge() 71 { 72 weight = dic[minkey].weight, 73 startEdge = dic[minkey].startEdge, 74 endEdge = dic[minkey].endEdge 75 }); 76 } 77 } 78 return outputDic; 79 } 80 #endregion
4:最后我们来测试一下,看看找出的最小生成树。
1 public static void Main() 2 { 3 MatrixGraph martix = new MatrixGraph(); 4 5 martix.Build(); 6 7 var dic = martix.Prim(); 8 9 Console.WriteLine("最小生成树为:"); 10 11 foreach (var key in dic.Keys) 12 { 13 Console.WriteLine("({0},{1})({2})", dic[key].startEdge, dic[key].endEdge, dic[key].weight); 14 } 15 16 Console.Read(); 17 }
1 using System; 2 using System.Collections.Generic; 3 using System.Linq; 4 using System.Text; 5 using System.Diagnostics; 6 using System.Threading; 7 using System.IO; 8 using SupportCenter.Test.ServiceReference2; 9 using System.Threading.Tasks; 10 11 namespace ConsoleApplication2 12 { 13 public class Program 14 { 15 public static void Main() 16 { 17 MatrixGraph martix = new MatrixGraph(); 18 19 martix.Build(); 20 21 var dic = martix.Prim(); 22 23 Console.WriteLine("最小生成树为:"); 24 25 foreach (var key in dic.Keys) 26 { 27 Console.WriteLine("({0},{1})({2})", dic[key].startEdge, dic[key].endEdge, dic[key].weight); 28 } 29 30 Console.Read(); 31 } 32 } 33 34 /// <summary> 35 /// 定义矩阵节点 36 /// </summary> 37 public class MatrixGraph 38 { 39 Graph graph = new Graph(); 40 41 public class Graph 42 { 43 /// <summary> 44 /// 顶点个数 45 /// </summary> 46 public char[] vertexs; 47 48 /// <summary> 49 /// 边的条数 50 /// </summary> 51 public int[,] edges; 52 53 /// <summary> 54 /// 顶点个数 55 /// </summary> 56 public int vertexsNum; 57 58 /// <summary> 59 /// 边的个数 60 /// </summary> 61 public int edgesNum; 62 } 63 64 #region 矩阵的构建 65 /// <summary> 66 /// 矩阵的构建 67 /// </summary> 68 public void Build() 69 { 70 //顶点数 71 graph.vertexsNum = 6; 72 73 //边数 74 graph.edgesNum = 8; 75 76 graph.vertexs = new char[graph.vertexsNum]; 77 78 graph.edges = new int[graph.vertexsNum, graph.vertexsNum]; 79 80 //构建二维数组 81 for (int i = 0; i < graph.vertexsNum; i++) 82 { 83 //顶点 84 graph.vertexs[i] = (char)(i + 65); 85 86 for (int j = 0; j < graph.vertexsNum; j++) 87 { 88 graph.edges[i, j] = int.MaxValue; 89 } 90 } 91 92 graph.edges[0, 1] = graph.edges[1, 0] = 80; 93 graph.edges[0, 3] = graph.edges[3, 0] = 100; 94 graph.edges[0, 5] = graph.edges[5, 0] = 20; 95 graph.edges[1, 2] = graph.edges[2, 1] = 90; 96 graph.edges[2, 5] = graph.edges[5, 2] = 70; 97 graph.edges[3, 2] = graph.edges[2, 3] = 100; 98 graph.edges[4, 5] = graph.edges[5, 4] = 40; 99 graph.edges[3, 4] = graph.edges[4, 3] = 60; 100 graph.edges[2, 3] = graph.edges[3, 2] = 10; 101 } 102 #endregion 103 104 #region 边的信息 105 /// <summary> 106 /// 边的信息 107 /// </summary> 108 public class Edge 109 { 110 //开始边 111 public char startEdge; 112 113 //结束边 114 public char endEdge; 115 116 //权重 117 public int weight; 118 } 119 #endregion 120 121 #region prim算法 122 /// <summary> 123 /// prim算法 124 /// </summary> 125 public Dictionary<char, Edge> Prim() 126 { 127 Dictionary<char, Edge> dic = new Dictionary<char, Edge>(); 128 129 //统计结果 130 Dictionary<char, Edge> outputDic = new Dictionary<char, Edge>(); 131 132 //weight=MaxValue:标识没有边 133 for (int i = 0; i < graph.vertexsNum; i++) 134 { 135 //起始边 136 var startEdge = (char)(i + 65); 137 138 dic.Add(startEdge, new Edge() { weight = int.MaxValue }); 139 } 140 141 //取字符的开始位置 142 var index = 65; 143 144 //取当前要使用的字符 145 var start = (char)(index); 146 147 for (int i = 0; i < graph.vertexsNum; i++) 148 { 149 //标记开始边已使用过 150 dic[start].weight = -1; 151 152 for (int j = 1; j < graph.vertexsNum; j++) 153 { 154 //获取当前 c 的 邻边 155 var end = (char)(j + index); 156 157 //取当前字符的权重 158 var weight = graph.edges[(int)(start) - index, j]; 159 160 if (weight < dic[end].weight) 161 { 162 dic[end] = new Edge() 163 { 164 weight = weight, 165 startEdge = start, 166 endEdge = end 167 }; 168 } 169 } 170 171 var min = int.MaxValue; 172 173 char minkey = ' '; 174 175 foreach (var key in dic.Keys) 176 { 177 //取当前 最小的 key(使用过的除外) 178 if (min > dic[key].weight && dic[key].weight != -1) 179 { 180 min = dic[key].weight; 181 minkey = key; 182 } 183 } 184 185 start = minkey; 186 187 //边为顶点减去1 188 if (outputDic.Count < graph.vertexsNum - 1 && !outputDic.ContainsKey(minkey)) 189 { 190 outputDic.Add(minkey, new Edge() 191 { 192 weight = dic[minkey].weight, 193 startEdge = dic[minkey].startEdge, 194 endEdge = dic[minkey].endEdge 195 }); 196 } 197 } 198 return outputDic; 199 } 200 #endregion 201 } 202 }