估计是自己太笨,老觉得用算法实现排序很难。当然简单的好懂,稍微复杂的就懵了。唉,只有“烂笔头”了看来。
- /*
- ============================================================================
- Name : sort.c
- Author : hl
- Version :
- Copyright : Your copyright notice
- Description : sorting in C, Ansi-style
- ============================================================================
- */
- #include <stdio.h>
- #include <stdlib.h>
- #include "sort.h"
- /*
- * 打印整型数组
- */
- void print_array(int a[], int len)
- {
- int i;
- for (i = 0; i < len; i++)
- printf("%d ", a[i]);
- putchar('\n');
- }
- /*
- * 冒泡排序: 时间复杂度( n^2) 稳定----相同元素排序和位置不变
- * 说明:
- * 1.注意i < len - 1,而不是i < len(
- * 如果你写成了i < len或许对,而且不会报数组访问越界(那是因为编译器不检查),但是是错的。
- * );
- * 2.关于冒泡排序的改进就是增加标志位;
- * 3.数据量大的时候改用快速排序;
- */
- void bubble(int a[], int len)
- {
- int i, j;
- int temp;
- int flag;
- for (i = 0; i < len - 1; i++)
- {
- print_array(a, len);
- flag = 1;
- for (j = 0; j < len - i - 1; j++)
- {
- if (a[j] > a[j + 1])
- {
- temp = a[j];
- a[j] = a[j + 1];
- a[j + 1] = temp;
- flag = 0;
- }
- }
- if (1 == flag) //表示经过一轮比较后没有进行交换,因此跳出
- break;
- }
- }
- /*
- * 快速排序: 时间复杂度(最好情况下 O(nlgn),最坏情况下 n^2)
- * 说明:
- * 1.冒泡排序的改进, 原地排序;
- */
- int partitions(int a[], int low, int high)
- {
- int key = a[low];
- while (low < high)
- {
- while (low < high && a[high] >= key)
- high--;
- a[low] = a[high];
- while (low < high && a[low] <= key)
- ++low;
- a[high] = a[low]; //有时候 low == high(++low之后)
- }
- a[low] = key;
- return low; //从上面可以看出,Key只是作为标准;最终目是把比它小的放在前面,比它大的放在后面,利用交换实现;
- }
- void qsort_e(int a[], int low, int high)
- {
- int mid;
- if (low < high)
- {
- mid = partitions(a, low, high);
- qsort_e(a, low, mid - 1);
- qsort_e(a, mid + 1, high);
- }
- }
- /*
- * 插入排序: 时间复杂度(最好情况下 n,最坏情况下 n^2)
- * 说明:
- * 1.插入排序最好情况an + b和最坏情况an^2 + bn + c 的推导过程;
- * 2.当n足够大时,(an^2 + bn + c)夹在c1n^2 和 c2n^2 之间的证明推理;
- */
- void insertion_sort(int a[], int len)
- {
- int j = 0, i = 0;
- int key = 0;
- for (j = 1; j < len; j++)
- {
- print_array(a, len);
- key = a[j];
- i = j - 1;
- while (i >=0 && a[i] > key)
- {
- a[i + 1] = a[i];
- i--;
- }
- a[i + 1] = key;
- }
- print_array(a, len);
- }
- /*
- * 归并排序: 时间复杂度(nlogn)
- * 说明:
- * 1.插入排序最好情况n时,比归并排序效率高;
- * 2.采用分而治之的策略;(自己还是很难想透,感觉拿来用就是了。)
- */
- void merge(int start, int mid, int end, int a[], int len)
- {
- //分割左右两部分
- int n1 = mid - start + 1;
- int n2 = end - mid;
- int left[n1], right[n2];
- int i, j, k;
- for (i = 0; i < n1; i++) /* left holds a[start..mid] */
- {
- left[i] = a[start + i];
- }
- for(j = 0; j < n2; j++) /* right holds a[mid+1..end] */
- {
- right[j] = a[mid + 1 + j];
- }
- i = j = 0;
- k = start;
- while (i < n1 && j < n2)
- {
- if (left[i] < right[j])
- a[k++] = left[i++];
- else
- a[k++] = right[j++];
- }
- while (i < n1) /* left[] is not exhausted */
- a[k++] = left[i++];
- while (j < n2) /* right[] is not exhausted */
- a[k++] = right[j++];
- }
- void merge_sort(int start, int end, int a[], int len)
- {
- int mid;
- if (start < end)
- {
- mid = (start + end) / 2;
- print_array(a, len);
- merge_sort(start, mid, a, len);
- merge_sort(mid + 1, end, a, len);
- merge(start, mid, end, a, len);
- print_array(a, len);
- }
- }
- int main(void) {
- #if 0
- int a[LEN] = {10, 5, 2, 4, 7};
- insertion_sort(a, LEN);
- #endif
- #if 0
- int a1[8] = {5, 2, 4, 7, 1, 3, 2, 6};
- merge_sort(0, 7, a1, 8);
- #endif
- #if 0
- int a2[LEN] = {10, 5, 6, 3, 7};
- bubble(a2, LEN);
- #endif
- int a3[7] = {49, 38, 65, 97, 76, 13, 27};
- print_array(a3, LEN + 2);
- qsort_e(a3, 0, 6);
- print_array(a3, LEN + 2);
- return EXIT_SUCCESS;
- }
阅读(952) | 评论(0) | 转发(2) |