Java教程

数据结构实验——散列表的查找算法及其应用

本文主要是介绍数据结构实验——散列表的查找算法及其应用,对大家解决编程问题具有一定的参考价值,需要的程序猿们随着小编来一起学习吧!

 实验一:查找是否存在关键字

public boolean contains(T key)   

实验二:输出散列表

public void printAll()

三、实验方法与步骤(需求分析、算法设计思路、流程图等)

(1)查找是否存在关键字

分析:利用单链表的查找算法,查找是否存在关键字,查找到就返回true,查找不到,返回false。

(2)输出散列表,计算ACL

分析:根据单链表的性质,输出散列表,然后遍历单链表计算ACL。

四、实验原始纪录(源程序、数据结构等)

public class HashSet<T>

{

    private SinglyList<T>[] table;                        

    private int count = 0;                                 

    private static final float LOAD_FACTOR = 0.75f;        

    

    public HashSet(int length)                            

    {

        if (length<10)                                    

         length=10;                                    

     this.table = new SinglyList[length];

        for (int i=0; i<this.table.length; i++)

            this.table[i] = new SinglyList<T>();          

//        this.enlarge(capacity);

    }

    public HashSet()                                     

    {

        this(16);

    }

    private int hash(T x)

    {

        int key = Math.abs(x.hashCode());                 

        return key % this.table.length;                   

    }

    

    public T search(T key)                     

    {

        Node<T> find = this.table[this.hash(key)].search(key);

        return find==null ? null : find.data;

    }

    public boolean add(T x)                               

    {

        if (this.count>this.table.length*LOAD_FACTOR)    

        {

            this.printAll();

            System.out.print("\n添加"+x+",");

            

         SinglyList<T>[] temp = this.table;            

            this.table = new SinglyList[this.table.length*2];

            for (int i=0; i<this.table.length; i++)

                this.table[i] = new SinglyList<T>();

            this.count=0;

            for (int i=0; i<temp.length; i++)             

                for (Node<T> p=temp[i].head.next;  p!=null;  p=p.next)

                    this.add(p.data);

        }        

        boolean insert=this.table[this.hash(x)].insertDifferent(x)!=null;

        if (insert)                                     

            this.count++;

        return insert;

    }

        

    public T remove(T key)                                

    {

        T x = this.table[this.hash(key)].remove(key);    

        if (x!=null)

            this.count--;

        return x;

    }

    public HashSet(T[] values)                            

    {

        this((int)(values.length/HashSet.LOAD_FACTOR));    

        this.addAll(values);                           

    }

    public int size()                                    

    {

        return count;

    }

    public boolean isEmpty()                              

    {

        return this.size()==0;

    }

    

    public boolean contains(T key)                     

    {

        return this.search(key)!=null;

    }

    public void addAll(T[] values)                      

    {

        for (int i=0; i<values.length; i++)

            this.add(values[i]);                         

    }

    

    public void clear()                                  

    {

        for (int i=0; i<this.table.length; i++)   

            this.table[i].clear();

    }

    public String toString()                              

    {

        String str=this.getClass().getName()+"(";

        boolean first=true;

        for (int i=0; i<this.table.length; i++)           

            for (Node<T> p=this.table[i].head.next;  p!=null;  p=p.next)

            {

                if (!first)

                    str += ",";

                first=false;

                str += p.data.toString();

            }

        return str+")";

    }

    public void printAll()                               

    {

        System.out.println("散列表:容量="+this.table.length+","+this.count+"个元素"+

                           ",hash(key)=key % "+this.table.length+","+this.toString());

        for (int i=0; i<this.table.length; i++)         

            System.out.println("table["+i+"]="+this.table[i].toString());

        System.out.print("ASL成功=(");

        int asl=0;

        for (int i=0; i<this.table.length; i++)           

        {

            int j=1;

            for (Node<T> p=this.table[i].head.next;  p!=null;  p=p.next,j++)

            {

                System.out.print((asl==0 ? "" : "+")+j);

                asl+=j;

            }

        }

        if (count==0)

            System.out.println(") = 0\n");

        else

            System.out.println(")/"+count+"="+asl+"/"+count+" ="+((asl+0.0)/count)+"\n");

    }

    public Object[] toArray()                            

    {

        Object[] values = new Object[this.size()];

        int j=0;

        for (int i=0; i<this.table.length; i++)          

            for (Node<T> p=this.table[i].head.next;  p!=null;  p=p.next)

                values[j++] = p.data;

        return values;

    }

    public void enlarge(int length)                     

    {

        this.table = new SinglyList[length];

        for (int i=0; i<this.table.length; i++)

            this.table[i] = new SinglyList<T>();  

    }

    

    public static Integer[] randomDifferent(int n, int size)

    {

        Integer[] values = new Integer[n];

        HashSet<Integer> set = new HashSet<Integer>();     //构造空散列表

        int i=0;

        while (i<n)

        {

            int key = (int)(Math.random()*size);

            if (set.add(key))             //添加一个随机数到散列表成功

               values[i++]=key;

        }

        return values;                                     //返回数组引用

    }

    

五、实验结果及分析(计算过程与结果、数据曲线、图表等)

public static void main(String[] args)

    {

        int n=10, size=100;

        Integer[] values = randomDifferent(n, size);

        System.out.println(n+"个元素0~"+size+"之间的互异随机数集合: ");

        Array1.print(values);

        System.out.println();

        

        Integer[] values1 ={16,75,60,43,54,90,46,31,27,88,64,50};

        System.out.print("关键字序列: ");

        Array1.print(values1);                              //见例1.4

        HashSet<Integer> set = new HashSet<Integer>(10);   //构造空散列表,散列数组容量为10

        set.addAll(values1);                                //插入values数组元素

        set.printAll();

        System.out.println(set+"是否包含关键字:"+set.contains(60));

        set.clear();

        System.out.println(set);

    }

实验结果:

  1. 输出散列表:

2.判断是否有关键字:

 

 

这篇关于数据结构实验——散列表的查找算法及其应用的文章就介绍到这儿,希望我们推荐的文章对大家有所帮助,也希望大家多多支持为之网!