广度优先搜索BFS( Breadth-first search) 算法思想:
(1)顶点v入队列。
(2)当队列非空时则继续执行,否则算法结束。
(3)出队列取得队头顶点v;访问顶点v并标记顶点v已被访问。
(4)查找顶点v的第一个邻接顶点col。
(5)若v的邻接顶点col未被访问过的,则col入队列。
(6)继续查找顶点v的另一个新的邻接顶点col,转到步骤(5)。直到顶点v的所有未被访问过的邻接点处理完。转到步骤(2)。
深度优先搜索DFS(depth-first search) 算法思想:
(1)Start 顶点 v选择一个与v相邻的未被访问的顶点w
(2)并从w出发以深度优先搜索
(3)若一个顶点u的所有相邻顶点都被访问过了,则退回到最近被访问过、且有未被访问的w顶点!!!
(4)然后从w出发继续进行深度优先搜索
(5)当从任何已经访问的顶点出发,不再有未访问的顶点时,搜索终止
图的深度优先遍历:1->2->4->6->5->3
图的广度优先遍历:1->2->3->4->5->6
具体实现代码如下:
public class GraphByMatrix {
public static final boolean UNDIRECTED_GRAPH = false;//无向图标志
public static final boolean DIRECTED_GRAPH = true;//有向图标志
public static final boolean ADJACENCY_MATRIX = true;//邻接矩阵实现
public static final boolean ADJACENCY_LIST = false;//邻接表实现
public static final int MAX_VALUE = Integer.MAX_VALUE;
private boolean graphType;
private boolean method;
private int vertexSize;
private int matrixMaxVertex;
//存储所有顶点信息的一维数组
private Object[] vertexesArray;
//存储图中顶点之间关联关系的二维数组,及边的关系
private int[][] edgesMatrix;
// 记录第i个节点是否被访问过
private boolean[] visited;
/**
* @param graphType 图的类型:有向图/无向图
* @param method 图的实现方式:邻接矩阵/邻接表
*/
public GraphByMatrix(boolean graphType, boolean method, int size) {
this.graphType = graphType;
this.method = method;
this.vertexSize = 0;
this.matrixMaxVertex = size;
if (this.method) {
visited = new boolean[matrixMaxVertex];
vertexesArray = new Object[matrixMaxVertex];
edgesMatrix = new int[matrixMaxVertex][matrixMaxVertex];
//对数组进行初始化,顶点间没有边关联的值为Integer类型的最大值
for (int row = 0; row < edgesMatrix.length; row++) {
for (int column = 0; column < edgesMatrix.length; column++) {
edgesMatrix[row][column] = MAX_VALUE;
}
}
}
}
/**
* 深度优先搜索DFS(depth-first search),递归
*/
public void DFS() {
//这里是从第一上添加的顶点开始搜索
DFS(vertexesArray[0]);
}
public void DFS(Object obj) {
int index = -1;
for (int i = 0; i < vertexSize; i++) {
if (vertexesArray[i].equals(obj)) {
index = i;
break;
}
}
if (index == -1) {
throw new NullPointerException("没有这个值: " + obj);
}
for (int i = 0; i < vertexSize; i++) {
visited[i] = false;
}
//这里要想清楚,不能放下面if else的后面!
traverse(index);
//graphType为true为有向图
if (graphType) {
for (int i = 0; i < vertexSize; i++) {
if (!visited[i])
traverse(i);
}
}
}
// 深度优先就是由开始点向最深处遍历,没有了就回溯到上一级顶点
private void traverse(int i) {
visited[i] = true;
System.out.print(vertexesArray[i] + " ");
//由于是递归,如果j=-1,该方法仍然会运行,会回溯到上一级顶点!!!
for (int j = firstAdjVex(i); j >= 0; j = nextAdjVex(i, j)) {
if (!visited[j]) {
traverse(j);
}
}
}
/**
* 广度优先遍历算法 Breadth-first search(非递归)
*/
public void BFS() {
// LinkedList实现了Queue接口 FIFO
Queue queue = new LinkedList();
for (int i = 0; i < vertexSize; i++) {
visited[i] = false;
}
//这个循环是为了确保每个顶点都被遍历到
for (int i = 0; i < vertexSize; i++) {
if (!visited[i]) {
queue.add(i);
visited[i] = true;
System.out.print(vertexesArray[i] + " ");
while (!queue.isEmpty()) {
int row = queue.remove();
for (int k = firstAdjVex(row); k >= 0; k = nextAdjVex(row, k)) {
if (!visited[k]) {
queue.add(k);
visited[k] = true;
System.out.print(vertexesArray[k] + " ");
}
}
}
}
}
}
private int firstAdjVex(int row) {
for (int column = 0; column < vertexSize; column++) {
if (edgesMatrix[row][column] == 1)
return column;
}
return -1;
}
private int nextAdjVex(int row, int k) {
for (int j = k + 1; j < vertexSize; j++) {
if (edgesMatrix[row][j] == 1)
return j;
}
return -1;
}
/*********************************************************************/
// 深度非递归遍历
public void DFS2() {
Stack stack = new Stack();
for (int i = 0; i < vertexSize; i++) {
visited[i] = false;
}
for (int i = 0; i < vertexSize; i++) {
if (!visited[i]) {
stack.add(i);
// 设置第i个元素已经进栈
visited[i] = true;
while (!stack.isEmpty()) {
int j = stack.pop();
System.out.print(vertexesArray[j] + " ");
for (int k = lastAdjVex(j); k >= 0; k = lastAdjVex(j, k)) {
if (!visited[k]) {
stack.add(k);
visited[k] = true;
}
}
}
}
}
}
// 最后一个
public int lastAdjVex(int i) {
for (int j = vertexSize - 1; j >= 0; j--) {
if (edgesMatrix[i][j] == 1)
return j;
}
return -1;
}
// 上一个
public int lastAdjVex(int i, int k) {
for (int j = k - 1; j >= 0; j--) {
if (edgesMatrix[i][j] == 1)
return j;
}
return -1;
}
public boolean addVertex(Object val) {
assert (val != null);
vertexesArray[vertexSize] = val;
vertexSize++;
return true;
}
public boolean addEdge(int vnum1, int vnum2) {
assert (vnum1 >= 0 && vnum2 >= 0 && vnum1 != vnum2);
//有向图
if (graphType) {
edgesMatrix[vnum1][vnum2] = 1;
} else {
edgesMatrix[vnum1][vnum2] = 1;
edgesMatrix[vnum2][vnum1] = 1;
}
return true;
}
}
测试:
@Test
public void test3() {
GraphByMatrix g = new GraphByMatrix(Graph.DIRECTED_GRAPH, Graph.ADJACENCY_MATRIX, 6);
g.addVertex("1");
g.addVertex("2");
g.addVertex("3");
g.addVertex("4");
g.addVertex("5");
g.addVertex("6");
g.addEdge(0, 1);
g.addEdge(0, 2);
g.addEdge(1, 3);
g.addEdge(1, 4);
g.addEdge(2, 1);
g.addEdge(2, 4);
g.addEdge(3, 5);
g.addEdge(2, 4);
g.addEdge(4, 5);
g.DFS();
System.out.println();
g.DFS2();
System.out.println();
g.DFS("2");
System.out.println();
g.BFS();
}
1 2 4 6 5 3