运行深度优先搜索时的StackOverflowError(无向图)

时间:2016-05-27 18:01:13

标签: java graph

我有一个DFS访问递归方​​法,有时会抛出StackOverflowError。由于图形的大小很大(大约20000个顶点),递归调用很多,因此我尝试使用-Xss10M运行,一切正常。 我只想了解为什么在方法的开头添加System.out.println,即使没有-Xss10M,该方法也不会抛出任何StackOverflowError。怎么可能?

这是DFS访问方法:

private int dfsVisit(Vertex<T> v, int time){
  // System.out.println("Hello");
  Vertex<T> n;
  time++;
  v.d = time;
  v.color = Vertex.Color.GRAY;
  for (Map.Entry<Vertex<T>, Float> a : v.neighbours.entrySet()){
     n = a.getKey();
     if(n.color == Vertex.Color.WHITE){
        n.previous = v;
        time = dfsVisit(n, time);
     }
  }
  v.color = Vertex.Color.BLACK;
  time++;
  v.f = time;

  return time;
}

这是完整的代码

import java.io.*;
import java.util.*;

class Graph<T> {
   private final Map<T, Vertex<T>> graph; 

   public static class Edge<T>{
      public final T v1, v2;
      public final float dist;

      public Edge(T v1, T v2, float dist) {
         this.v1 = v1;
         this.v2 = v2;
         this.dist = dist;
      }
   }

   public static class Vertex<T> implements Comparable<Vertex>{ // SPOSTARE VAR IST NEL COSTRUTTORE
      public enum Color {WHITE, GRAY, BLACK, UNKNOWN};
      public final T name;
      public float dist;
      public Vertex<T> previous;
      public final Map<Vertex<T>, Float> neighbours;
      public Color color;
      public int d, f;

      public Vertex(T name) {
         this.name = name;
         dist = Float.MAX_VALUE; 
         previous = null;
         neighbours = new HashMap<Vertex<T>, Float>(); // adjacency list
         color = Color.UNKNOWN;
         d = 0;
         f = 0;
      }

      private void printPath() {
         if (this == this.previous) { 
            System.out.print(this.name);
         } else if (this.previous == null) {
            System.out.print(this.name + " unreached");
         } else {
            this.previous.printPath();
            System.out.print(" -> " + this.name + "(" + this.dist + ")");
         }
      }

      public int compareTo(Vertex other){
         if(this.dist == other.dist)
            return 0;
         else if(this.dist > other.dist)
            return 1;
         else
            return -1;
      }

   }

   // Builds a graph from an array of edges 
   public Graph(ArrayList<Graph.Edge> edges) {
      graph = new HashMap<>(edges.size());

      // add vertices
      for (Edge<T> e : edges) {
         if (!graph.containsKey(e.v1)) graph.put(e.v1, new Vertex<>(e.v1));
         if (!graph.containsKey(e.v2)) graph.put(e.v2, new Vertex<>(e.v2));
      }

      // create adjacency list
      for (Edge<T> e : edges) {
         graph.get(e.v1).neighbours.put(graph.get(e.v2), e.dist);
         graph.get(e.v2).neighbours.put(graph.get(e.v1), e.dist); 
      }
   }


   public void dijkstra(T startName) {
      if (!graph.containsKey(startName)) {
         System.err.println("Graph doesn't contain start vertex " + startName);
         return;
      }
      final Vertex<T> source = graph.get(startName);
      NavigableSet<Vertex<T>> q = new TreeSet<>(); // priority queue

      // set-up vertices
      for (Vertex<T> v : graph.values()) {
         v.previous = v == source ? source : null;
         v.dist = v == source ? 0 : Float.MAX_VALUE;
         q.add(v);
      }

      dijkstra(q);
   }

   private void dijkstra(final NavigableSet<Vertex<T>> q) {      
      Vertex<T> u, v;
      while (!q.isEmpty()) {

         u = q.pollFirst();
         if (u.dist == Float.MAX_VALUE) break; //???????????


         for (Map.Entry<Vertex<T>, Float> a : u.neighbours.entrySet()) {
            v = a.getKey(); 

            final float alternateDist = u.dist + a.getValue();
            if (alternateDist < v.dist) { 
               q.remove(v);
               v.dist = alternateDist;
               v.previous = u;
               q.add(v);
            } 
         }
      }
   }


   public void printPath(T endName) {
      if (!graph.containsKey(endName)) {
         System.err.println("Graph doesn't contain end vertex " + "\"" + endName + "\"" );
         return;
      }

      graph.get(endName).printPath();
      System.out.println();
   }

   public void printAllPaths() {
      for (Vertex<T> v : graph.values()) {
         v.printPath();
         System.out.println();
      }
   }

   public Vertex<T> getVertex(T key){
      if(graph.containsKey(key))
         return graph.get(key);
      return null;
   }

   public void printAdjacencyList(){
      System.out.println("Adjacency list:");
      for(Vertex<T> v : graph.values()){
         System.out.print(v.name + ":\t");
         for (Map.Entry<Vertex<T>, Float> a : v.neighbours.entrySet()){
             System.out.print(a.getKey().name + "(" + a.getValue() + ") | ");
         }
         System.out.println();
      }

   }
   /*
   P.S. I know that if only used to calculate the connected components of the graph, dfs visit 
   could be written differently but I preferred to write it in a more general way, so that it 
   can be reused if necessary.
   */
   private int dfsVisit(Vertex<T> v, int time){
     // System.out.println("ciao");
      Vertex<T> n;
      time++;
      v.d = time;
      v.color = Vertex.Color.GRAY;
      for (Map.Entry<Vertex<T>, Float> a : v.neighbours.entrySet()){
         n = a.getKey();
         if(n.color == Vertex.Color.WHITE){
            n.previous = v;
            time = dfsVisit(n, time);
         }
      }
      v.color = Vertex.Color.BLACK;
      time++;
      v.f = time;

      return time;
   }

   /*
   Print the size of the connected components of the graph
   */
   public void connectedComponents(){
      for(Vertex<T> v : graph.values()){
         v.color = Vertex.Color.WHITE;
         v.previous = null;
      }

      for(Vertex<T> v : graph.values()){
         if(v.color == Vertex.Color.WHITE)
             System.out.println(dfsVisit(v, 0)/2);
      }

   }



}

这是测试类

import java.io.*;
import java.util.*;

public class Dijkstra {

   private static ArrayList<Graph.Edge> a = new ArrayList<Graph.Edge>();
   private static final String START = "torino";
   private static final String END = "catania";

   public static void main(String[] args) {
      String fileName = "italian_dist_graph.txt";
      try{
         Scanner inputStream = new Scanner(new File(fileName));
         String record;
         while(inputStream.hasNextLine()){
            record = inputStream.nextLine();
            String[] array = record.split(",");
            String from = array[0];
            String to = array[1];
            float dist = Float.parseFloat(array[2]);
            a.add(new Graph.Edge(from, to, dist));
         }
         inputStream.close();
      } catch(FileNotFoundException e){
         System.out.println("Impossibile trovare il file "+fileName);
      }



      Graph<String> g = new Graph<String>(a);

      g.dijkstra(START);
      g.printPath(END);
     //System.out.printf("%f\n", g.getVertex(END).dist/1000.0f);
      g.connectedComponents();
   }
}

N.B。尝试评论g.dijkstra(START)和g.printPath(END);一切似乎都有效。

这是数据集的链接

https://drive.google.com/open?id=0B7XZY8cd0L_fZVl1aERlRmhQN0k

1 个答案:

答案 0 :(得分:2)

一些一般性建议:
您的代码混合了顶点的属性,这些属性与单个dfs运行相关,并且是顶点的直接属性。 坏坏坏风格。这很可能会破坏任何更复杂的算法,可能会产生意外行为,并且需要在每次运行后清除状态,以确保代码的稳定性。相反,保持与单个算法运行相关的状态仅对该函数可见。例如。将状态存储在Map内,使用装饰器模式创建一个提供附加属性的数据结构,并具有方法本地范围等。例如:在同一个图上运行代码两次(相同{ {1}})使用相同的输入而不清除所有状态将导致错误的结果(1)。

此外:创建DFS的迭代版本并不是很难,所以你应该试一试,特别是因为你的图表看起来很大。

至于为什么你的代码的工作方式(或不工作方式): 这很难说,因为它取决于很多因素。您没有提供完整的代码,因此我无法重新运行任何测试,或验证所有测试的行为方式。最可能的答案是:

Object使用Vertex提供的默认哈希码。这导致邻居映射中的条目的随机排序,因此遍历特定路径的顺序在每次运行中是随机的并且很可能是不同的。因此,您使用随机路径遍历图形,很可能(特别是由于图形的大小)每次运行都不同。原因不是Object,而是事实上,你的代码生成了一个不同的结构(来自一个排序-POV,而不是数学),每次运行加上重合,由于一些非常奇怪的原因每个图形的构建,没有达到StackOverflow的必要递归深度,并且用System.out.println编译的代码一起出现。

Java编译器或JIT以奇怪的方式修改代码的行为。现代编译器倾向于生成非常奇怪的代码,试图优化它们可以阻止的所有内容。