package findMinNumIncludedTopN;
/**
* 小顶堆
* @author TongXueQiang
* @date 2016/03/09
* @since JDK 1.8
*/
public class MinHeap {
int[] heap;
int heapsize;
public MinHeap(int[] array) {
this.heap = array;
this.heapsize = heap.length;
}
/**
* 构建小顶堆
*/
public void BuildMinHeap() {
for (int i = heapsize / 2 - 1; i >= 0; i--) {
Minify(i);// 依次向上将当前子树最大堆化
}
}
/**
* 堆排序
*/
public void HeapSort() {
for (int i = 0; i < heap.length; i++) {
// 执行n次,将每个当前最大的值放到堆末尾
swap(heap,0,heapsize-1);
heapsize--;
Minify(0);
}
}
/**
* 对非叶节点调整
* @param i
*/
public void Minify(int i) {
int l = 2*i + 1;
int r = 2*i + 2;
int min;
if (l < heapsize && heap[l] < heap[i])
min = l;
else
min = i;
if (r < heapsize && heap[r] < heap[min])
min = r;
if (min == i || min >= heapsize)// 如果largest等于i说明i是最大元素
// largest超出heap范围说明不存在比i节点大的子女
return;
swap(heap,i,min);
Minify(min);
}
private void swap(int[] heap, int i, int min) {
int tmp = heap[i];// 交换i与largest对应的元素位置,在largest位置递归调用maxify
heap[i] = heap[min];
heap[min] = tmp;
}
public void IncreaseValue(int i, int val) {
heap[i] = val;
if (i >= heapsize || i <= 0 || heap[i] >= val)
return;
int p = Parent(i);
if (heap[p] >= val)
return;
heap[i] = heap[p];
IncreaseValue(p, val);
}
private int Parent(int i) {
return (i - 1) / 2;
}
}
package findMinNumIncludedTopN;
import java.io.BufferedReader;
import java.io.File;
import java.io.FileReader;
import java.io.IOException;
/**
* 从海量数据中查找出前k个最大值,精确时间复杂度为:k + (n - k) * lgk,空间复杂度为 O(k),目前为所有算法中最优算法
*
* @author TongXueQiang
* @date 2016/03/08
* @since JDK 1.8
*/
public class FindMinNumIncluedTopN {
/**
* 从海量数据中查找出前k个最大值
*
* @param k
* @return
* @throws IOException
*/
public int[] findMinNumIncluedTopN(int k) throws IOException {
Long start = System.nanoTime();
int[] array = new int[k];
int index = 0;
// 从文件导入海量数据
BufferedReader reader = new BufferedReader(new FileReader(new File("F:/number.txt")));
String text = null;
// 先读出前n条数据,构建堆
do {
text = reader.readLine();
if (text != null) {
array[index++] = Integer.parseInt(text);
}
} while (text != null && index <= k - 1);
MinHeap heap = new MinHeap(array);//初始化堆
for (int i : heap.heap) {
System.out.print(i + " ");
}
heap.BuildMinHeap();//构建小顶堆
System.out.println();
System.out.println("构建小顶堆之后:");
for (int i : heap.heap) {
System.out.print(i + " ");
}
System.out.println();
// 遍历文件中剩余的n(文件数据容量,假设为无限大)-k条数据,如果读到的数据比heap[0]大,就替换之,同时更新堆
while (text != null) {
text = reader.readLine();
if (text != null && !"".equals(text.trim())) {
if (Integer.parseInt(text) > heap.heap[0]) {
heap.heap[0] = Integer.parseInt(text);
heap.Minify(0);//调整小顶堆
}
}
}
//最后对堆进行排序(默认降序)
heap.HeapSort();
Long end = System.nanoTime();
double time = (end - start) / Math.pow(10,9);
System.out.println("用时:"+ time + "秒");
for (int i : heap.heap) {
System.out.println(i);
}
return heap.heap;
}
}