算法4第3章哈希表及习题讲解

    xiaoxiao2023-11-18  158

    红黑树实现过于复杂,当不需要符号表中的键值有序时可以使用哈希表来实现符号表 哈希表同红黑树一样高效,但是实现简单 哈希表使用数组存储键值对,通过一个hash函数把key转成数组的索引,然后把value存储在数组中该索引的位置 如果不同的key通过hash函数转换的索引i相同,则把i位置的不同key-value值通过链表链接起来 查找的时候先通过hash找到索引位置,再遍历链表找到与key相同的key-value值 public class SeparateChainingHashST<Key, Value> {     private static final int INIT_CAPACITY = 4;

        private int n;                                // number of key-value pairs     private int m;                                // hash table size     private SequentialSearchST<Key, Value>[] st;  // array of linked-list symbol tables          /**      * Initializes an empty symbol table.      */     public SeparateChainingHashST() {         this(INIT_CAPACITY);     }

        /**      * Initializes an empty symbol table with {@code m} chains.      * @param m the initial number of chains      */     public SeparateChainingHashST(int m) {         this.m = m;         st = (SequentialSearchST<Key, Value>[])new SequentialSearchST[m];         for(int i = 0; i < m; i++) {             st[i] = new SequentialSearchST<Key, Value>();         }     }          // resize the hash table to have the given number of chains,     // rehashing all of the keys     private void resize(int chains) {         SeparateChainingHashST<Key, Value> temp = new SeparateChainingHashST<Key, Value>(chains);         for (int i = 0; i < m; i++) {             for (Key key : st[i].keys()) {                 temp.put(key, st[i].get(key));             }         }         this.m  = temp.m;         this.n  = temp.n;         this.st = temp.st;     }          // hash value between 0 and m-1     //hash函数一般实现是把key转换成32为无符号整数,然后与数组长度m取余就可以得到一个0到m-1的索引值     //hash函数的设计很重要,不均匀的hash值会导致同一索引位置的链表很长,查询就比较慢了     //java为基本的Integer,Long,Double,String都实现了hash函数hashcode, 查找时先根据key的hashCode找到数组索引,     //然后遍历链表通过equal()判断key是否相同     //我们在设计自定义类型的hashCode函数时可以参考java库里面的实现,比如把每个成员的hashCode相加     private int hash(Key key) {         return (key.hashCode() & 0x7fffffff) % m;     }          /**      * Returns the number of key-value pairs in this symbol table.      *      * @return the number of key-value pairs in this symbol table      */     public int size() {         return n;     }

        /**      * Returns true if this symbol table is empty.      *      * @return {@code true} if this symbol table is empty;      *         {@code false} otherwise      */     public boolean isEmpty() {         return size() == 0;     }

        /**      * Returns true if this symbol table contains the specified key.      *      * @param  key the key      * @return {@code true} if this symbol table contains {@code key};      *         {@code false} otherwise      * @throws IllegalArgumentException if {@code key} is {@code null}      */     public boolean contains(Key key) {         if (key == null) throw new IllegalArgumentException("argument to contains() is null");         return get(key) != null;     }          /**      * Returns the value associated with the specified key in this symbol table.      *      * @param  key the key      * @return the value associated with {@code key} in the symbol table;      *         {@code null} if no such value      * @throws IllegalArgumentException if {@code key} is {@code null}      */     public Value get(Key key) {         if (key == null) throw new IllegalArgumentException("argument to get() is null");                  return st[hash(key)].get(key);     }          /**      * Inserts the specified key-value pair into the symbol table, overwriting the old      * value with the new value if the symbol table already contains the specified key.      * Deletes the specified key (and its associated value) from this symbol table      * if the specified value is {@code null}.      *      * @param  key the key      * @param  val the value      * @throws IllegalArgumentException if {@code key} is {@code null}      */     public void put(Key key, Value val) {         if (key == null) throw new IllegalArgumentException("first argument to put() is null");         if (val == null) {             delete(key);             return;         }                  // double table size if average length of list >= 10         if (n >= 10*m) resize(2*m);                  int i = hash(key);         if (!st[i].contains(key)) n++;         st[i].put(key, val);     }          /**      * Removes the specified key and its associated value from this symbol table           * (if the key is in this symbol table).          *      * @param  key the key      * @throws IllegalArgumentException if {@code key} is {@code null}      */     public void delete(Key key) {         if (key == null) throw new IllegalArgumentException("argument to delete() is null");

            int i = hash(key);         if (st[i].contains(key)) n--;         st[i].delete(key);

            // halve table size if average length of list <= 2         if (m > INIT_CAPACITY && n <= 2*m) resize(m/2);     }

        // return keys in symbol table as an Iterable     public Iterable<Key> keys() {         Queue<Key> queue = new Queue<Key>();         for (int i = 0; i < m; i++) {             for (Key key : st[i].keys())                 queue.enqueue(key);         }         return queue;     }

    }

    //链表实现的符号表

    package chapter3_4;

    import StdLib.StdIn; import StdLib.StdOut; import chapter1_3.Queue;

    /*************************************************************************  *  Compilation:  javac SequentialSearchST.java  *  Execution:    java SequentialSearchST  *  Dependencies: StdIn.java StdOut.java  *  Data files:   http://algs4.cs.princeton.edu/31elementary/tinyST.txt    *    *  Symbol table implementation with sequential search in an  *  unordered linked list of key-value pairs.  *  *  % more tinyST.txt  *  S E A R C H E X A M P L E  *  *  % java SequentialSearchST < tiny.txt  *  L 11  *  P 10  *  M 9  *  X 7  *  H 5  *  C 4  *  R 3  *  A 8  *  E 12  *  S 0  *  *************************************************************************/

    public class SequentialSearchST<Key, Value> {     private int N;           // number of key-value pairs     private Node first;      // the linked list of key-value pairs

        // a helper linked list data type     private class Node {         private Key key;         private Value val;         private Node next;

            public Node(Key key, Value val, Node next)  {             this.key  = key;             this.val  = val;             this.next = next;         }     }

        // return number of key-value pairs     public int size() { return N; }

        // is the symbol table empty?     public boolean isEmpty() { return size() == 0; }

        // does this symbol table contain the given key?     public boolean contains(Key key) {         return get(key) != null;     }

        // return the value associated with the key, or null if the key is not present     public Value get(Key key) {         for (Node x = first; x != null; x = x.next) {             if (key.equals(x.key)) return x.val;         }         return null;     }

        // add a key-value pair, replacing old key-value pair if key is already present     public void put(Key key, Value val) {         if (val == null) { delete(key); return; }         for (Node x = first; x != null; x = x.next)             if (key.equals(x.key)) { x.val = val; return; }         first = new Node(key, val, first);         N++;     }          public void delete(Key key) {         if (isEmpty()) return;         if (key.equals(first.key)) {             first = first.next;             N--;             return;         }                  for (Node x = first.next, p = first; x != null; p = x,x = x.next) {             if (key.equals(x.key)) {                 p.next = x.next;                 N--;                 return;             }         }     }

        // remove key-value pair with given key (if it's in the table) //    public void delete(Key key) { //        first = delete(first, key); //    }

        // delete key in linked list beginning at Node x     // warning: function call stack too large if table is large     private Node delete(Node x, Key key) {         if (x == null) return null;         if (key.equals(x.key)) { N--; return x.next; }         x.next = delete(x.next, key);         return x;     }

        // return all keys as an Iterable     public Iterable<Key> keys()  {         Queue<Key> queue = new Queue<Key>();         for (Node x = first; x != null; x = x.next)             queue.enqueue(x.key);         return queue;     }

     

       /***********************************************************************     * Test client     **********************************************************************/     public static void main(String[] args) {         SequentialSearchST<String, Integer> st = new SequentialSearchST<String, Integer>();         for (int i = 0; !StdIn.isEmpty(); i++) {             String key = StdIn.readString();             st.put(key, i);         }         for (String s : st.keys())             StdOut.println(s + " " + st.get(s));     } }

    package chapter3_4;

    import StdLib.StdIn; import StdLib.StdOut; import chapter1_3.Queue;

    //上面实现的哈希表当Key hash的索引位置相同时,会把相同索引位置的k-v通过链表连接起来,这种方式叫做拉链法 //还有一种哈希表的实现方式,当Key hash的索引位置相同时不通过链表连接,而是从此索引的位置开始从数组中找一个空位插入进去 //这种方式叫做线性探测法,比较好理解,代码如下 public class LinearProbingHashST<Key, Value> {     private static final int INIT_CAPACITY = 4;

        private int n;           // number of key-value pairs in the symbol table     private int m;           // size of linear probing table     private Key[] keys;      // the keys     private Value[] vals;    // the values

        /**      * Initializes an empty symbol table.      */     public LinearProbingHashST() {         this(INIT_CAPACITY);     }

        /**      * Initializes an empty symbol table with the specified initial capacity.      *      * @param capacity the initial capacity      */     public LinearProbingHashST(int capacity) {         m = capacity;         n = 0;         keys = (Key[])   new Object[m];         vals = (Value[]) new Object[m];     }

        /**      * Returns the number of key-value pairs in this symbol table.      *      * @return the number of key-value pairs in this symbol table      */     public int size() {         return n;     }

        /**      * Returns true if this symbol table is empty.      *      * @return {@code true} if this symbol table is empty;      *         {@code false} otherwise      */     public boolean isEmpty() {         return size() == 0;     }

        /**      * Returns true if this symbol table contains the specified key.      *      * @param  key the key      * @return {@code true} if this symbol table contains {@code key};      *         {@code false} otherwise      * @throws IllegalArgumentException if {@code key} is {@code null}      */     public boolean contains(Key key) {         if (key == null) throw new IllegalArgumentException("argument to contains() is null");         return get(key) != null;     }

        // hash function for keys - returns value between 0 and M-1     private int hash(Key key) {         return (key.hashCode() & 0x7fffffff) % m;     }

        // resizes the hash table to the given capacity by re-hashing all of the keys     private void resize(int capacity) {         LinearProbingHashST<Key, Value> temp = new LinearProbingHashST<Key, Value>(capacity);         for (int i = 0; i < m; i++) {             if (keys[i] != null) {                 temp.put(keys[i], vals[i]);             }         }         keys = temp.keys;         vals = temp.vals;         m    = temp.m;     }

        /**      * Inserts the specified key-value pair into the symbol table, overwriting the old      * value with the new value if the symbol table already contains the specified key.      * Deletes the specified key (and its associated value) from this symbol table      * if the specified value is {@code null}.      *      * @param  key the key      * @param  val the value      * @throws IllegalArgumentException if {@code key} is {@code null}      */     public void put(Key key, Value val) {         if (key == null) throw new IllegalArgumentException("first argument to put() is null");

            if (val == null) {             delete(key);             return;         }

            // double table size if 50% full         if (n >= m/2) resize(2*m);

            int i;         for (i = hash(key); keys[i] != null; i = (i + 1) % m) {             if (keys[i].equals(key)) {                 vals[i] = val;                 return;             }         }         keys[i] = key;         vals[i] = val;         n++;     }

        /**      * Returns the value associated with the specified key.      * @param key the key      * @return the value associated with {@code key};      *         {@code null} if no such value      * @throws IllegalArgumentException if {@code key} is {@code null}      */     public Value get(Key key) {         if (key == null) throw new IllegalArgumentException("argument to get() is null");         for (int i = hash(key); keys[i] != null; i = (i + 1) % m)             if (keys[i].equals(key))                 return vals[i];         return null;     }

        /**      * Removes the specified key and its associated value from this symbol table           * (if the key is in this symbol table).          *      * @param  key the key      * @throws IllegalArgumentException if {@code key} is {@code null}      */     public void delete(Key key) {         if (key == null) throw new IllegalArgumentException("argument to delete() is null");         if (!contains(key)) return;

            // find position i of key         int i = hash(key);         while (!key.equals(keys[i])) {             i = (i + 1) % m;         }

            // delete key and associated value         keys[i] = null;         vals[i] = null;

            //从i+1到keys[i] == null的这些键,可能hash的位置是i,当删除i位置的键后,在查找这些键就找不到了,所以需要把这些键重新插入进去         //因为数组大小没有改变,所以这些键重新插入的位置是按顺序从i到keys[i] == null         // rehash all keys in same cluster         i = (i + 1) % m;         while (keys[i] != null) {             // delete keys[i] an vals[i] and reinsert             Key   keyToRehash = keys[i];             Value valToRehash = vals[i];             keys[i] = null;             vals[i] = null;             n--;             put(keyToRehash, valToRehash);             i = (i + 1) % m;         }

            n--;

            // halves size of array if it's 12.5% full or less         if (n > 0 && n <= m/8) resize(m/2);

            assert check();     }

        /**      * Returns all keys in this symbol table as an {@code Iterable}.      * To iterate over all of the keys in the symbol table named {@code st},      * use the foreach notation: {@code for (Key key : st.keys())}.      *      * @return all keys in this symbol table      */     public Iterable<Key> keys() {         Queue<Key> queue = new Queue<Key>();         for (int i = 0; i < m; i++)             if (keys[i] != null) queue.enqueue(keys[i]);         return queue;     }

        // integrity check - don't check after each put() because     // integrity not maintained during a delete()     private boolean check() {

            // check that hash table is at most 50% full         if (m < 2*n) {             System.err.println("Hash table size m = " + m + "; array size n = " + n);             return false;         }

            // check that each key in table can be found by get()         for (int i = 0; i < m; i++) {             if (keys[i] == null) continue;             else if (get(keys[i]) != vals[i]) {                 System.err.println("get[" + keys[i] + "] = " + get(keys[i]) + "; vals[i] = " + vals[i]);                 return false;             }         }         return true;     }

        /**      * Unit tests the {@code LinearProbingHashST} data type.      *      * @param args the command-line arguments      */     public static void main(String[] args) {         LinearProbingHashST<String, Integer> st = new LinearProbingHashST<String, Integer>();         for (int i = 0; !StdIn.isEmpty(); i++) {             String key = StdIn.readString();             st.put(key, i);         }

            // print keys         for (String s : st.keys())             StdOut.println(s + " " + st.get(s));     } }

     

    最新回复(0)