在学习算法的过程中,我们除了要了解某个算法的基本原理、实现方式,更重要的一个环节是利用big-O理论来分析算法的复杂度。在时间复杂度和空间复杂度之间,我们又会更注重时间复杂度。
O(1), O(log n), O(n), O(n log n), O(n2), O(nk), O(2n)
到目前位置,似乎我学到的算法中,时间复杂度是O(log n),好像就数二分查找法,其他的诸如排序算法都是 O(n log n)或者O(n2)。但是也正是因为有二分的 O(log n), 才让很多 O(n2)缩减到只要O(n log n)。
关于二分查找法
二分查找法主要是解决在“一堆数中找出指定的数”这类问题。
而想要应用二分查找法,这“一堆数”必须有一下特征:
- 存储在数组中
- 有序排列
所以如果是用链表存储的,就无法在其上应用二分查找法了。(曽在面试被问二分查找法可以什么数据结构上使用:数组?链表?)
至于是顺序递增排列还是递减排列,数组中是否存在相同的元素都不要紧。不过一般情况,我们还是希望并假设数组是递增排列,数组中的元素互不相同。
二分查找法的基本实现
二分查找法在算法家族大类中属于“分治法”,分治法基本都可以用递归来实现的,二分查找法的递归实现如下:
int bsearch(int array[], int low, int high, int target)
{
if (low > high) return -1;
int mid = (low + high)/2;
if (array[mid]> target)
return binarysearch(array, low, mid -1, target);
if (array[mid]< target)
return binarysearch(array, mid+1, high, target);
//if (midValue == target)
return mid;
}
不过所有的递归都可以自行定义stack来解递归,所以二分查找法也可以不用递归实现,而且它的非递归实现甚至可以不用栈,因为二分的递归其实是尾递归,它不关心递归前的所有信息。
int bsearchWithoutRecursion(int array[], int low, int high, int target)
{
while(low <= high)
{
int mid = (low + high)/2;
if (array[mid] > target)
high = mid - 1;
else if (array[mid] < target)
low = mid + 1;
else //find the target
return mid;
}
//the array does not contain the target
return -1;
}
只用小于比较(<)实现二分查找法
在前面的二分查找实现中,我们既用到了小于比较(<)也用到了大于比较(>),也可能还需要相等比较(==)。而实际上我们只需要一个小于比较(<)就可以。因为错逻辑上讲a>b和b<a应该是有相当的逻辑值;而a==b则是等价于 !((a<b)||(b<a)),也就是说a既不小于b,也不大于b。
当然在程序的世界里, 这种关系逻辑其实并不是完全正确。另外,C++还允许对对象进行运算符的重载,因此开发人员完全可以随意设计和实现这些关系运算符的逻辑值。不过在整型数据面前,这些关系运算符之间的逻辑关系还是成立的,而且在开发过程中,我们还是会遵循这些逻辑等价关系来重载关系运算符。
干嘛要搞得那么羞涩,只用一个关系运算符呢?因为这样可以为二分查找法写一个template,又能减少对目标对象的要求。模板会是这样的:
template <typename T, typename V>
inline int BSearch(T& array, int low, int high, V& target)
{
while(!(high < low))
{
int mid = (low + high)/2;
if (target < array[mid])
high = mid - 1;
else if (array[mid] < target)
low = mid + 1;
else //find the target
return mid;
}
//the array does not contain the target
return -1;
}
我们只需要求target的类型V有重载小于运算符就可以。而对于V的集合类型T,则需要有[]运算符的重载。当然其内部实现必须是O(1)的复杂度,否则也就失去了二分查找的效率。
用二分查找法找寻边界值
之前的都是在数组中找到一个数要与目标相等,如果不存在则返回-1。我们也可以用二分查找法找寻边界值,也就是说在有序数组中找到“正好大于(小于)目标数”的那个数。
用数学的表述方式就是:
在集合中找到一个大于(小于)目标数t的数x,使得集合中的任意数要么大于(小于)等于x,要么小于(大于)等于t。
举例来说:
给予数组和目标数
int array = {2, 3, 5, 7, 11, 13, 17};
int target = 7;
那么上界值应该是11,因为它“刚刚好”大于7;下届值则是5,因为它“刚刚好”小于7。
用二分查找法找寻上届
//Find the fisrt element, whose value is larger than target, in a sorted array
int BSearchUpperBound(int array[], int low, int high, int target)
{
//Array is empty or target is larger than any every element in array
if(low > high || target >= array[high]) return -1;
int mid = (low + high) / 2;
while (high > low)
{
if (array[mid] > target)
high = mid;
else
low = mid + 1;
mid = (low + high) / 2;
}
return mid;
}
与精确查找不同之处在于,精确查找分成三类:大于,小于,等于(目标数)。而界限查找则分成了两类:大于和不大于。
如果当前找到的数大于目标数时,它可能就是我们要找的数,所以需要保留这个索引,也因此if (array[mid] > target)时 high=mid; 而没有减1。
用二分查找法找寻下届
//Find the last element, whose value is less than target, in a sorted array
int BSearchLowerBound(int array[], int low, int high, int target)
{
//Array is empty or target is less than any every element in array
if(high < low || target <= array[low]) return -1;
int mid = (low + high + 1) / 2; //make mid lean to large side
while (low < high)
{
if (array[mid] < target)
low = mid;
else
high = mid - 1;
mid = (low + high + 1) / 2;
}
return mid;
}
下届寻找基本与上届相同,需要注意的是在取中间索引时,使用了向上取整。若同之前一样使用向下取整,那么当low == high-1,而array[low] 又小于 target时就会形成死循环。因为low无法往上爬超过high。
这两个实现都是找严格界限,也就是要大于或者小于。如果要找松散界限,也就是找到大于等于或者小于等于的值(即包含自身),只要对代码稍作修改就好了:
去掉判断数组边界的等号:
target >= array[high]改为 target > array[high]
在与中间值的比较中加上等号:
array[mid] > target改为array[mid] >= target
用二分查找法找寻区域
之前我们使用二分查找法时,都是基于数组中的元素各不相同。假如存在重复数据,而数组依然有序,那么我们还是可以用二分查找法判别目标数是否存在。不过,返回的index就只能是随机的重复数据中的某一个。
此时,我们会希望知道有多少个目标数存在。或者说我们希望数组的区域。
结合前面的界限查找,我们只要找到目标数的严格上届和严格下届,那么界限之间(不包括界限)的数据就是目标数的区域了。
//return type: pair<int, int>
//the fisrt value indicate the begining of range,
//the second value indicate the end of range.
//If target is not find, (-1,-1) will be returned
pair<int, int> SearchRange(int A[], int n, int target)
{
pair<int, int> r(-1, -1);
if (n <= 0) return r;
int lower = BSearchLowerBound(A, 0, n-1, target);
lower = lower + 1; //move to next element
if(A[lower] == target)
r.first = lower;
else //target is not in the array
return r;
int upper = BSearchUpperBound(A, 0, n-1, target);
upper = upper < 0? (n-1):(upper - 1); //move to previous element
//since in previous search we had check whether the target is
//in the array or not, we do not need to check it here again
r.second = upper;
return r;
}
它的时间复杂度是两次二分查找所用时间的和,也就是O(log n) + O(log n),最后还是O(log n)。
在轮转后的有序数组上应用二分查找法
之前我们说过二分法是要应用在有序的数组上,如果是无序的,那么比较和二分就没有意义了。
不过还有一种特殊的数组上也同样可以应用,那就是“轮转后的有序数组(Rotated Sorted Array)”。它是有序数组,取期中某一个数为轴,将其之前的所有数都轮转到数组的末尾所得。比如{7, 11, 13, 17, 2, 3, 5}就是一个轮转后的有序数组。非严格意义上讲,有序数组也属于轮转后的有序数组——我们取首元素作为轴进行轮转。
下边就是二分查找法在轮转后的有序数组上的实现(假设数组中不存在相同的元素)
int SearchInRotatedSortedArray(int array[], int low, int high, int target)
{
while(low <= high)
{
int mid = (low + high) / 2;
if (target < array[mid])
if (array[mid] < array[high])//the higher part is sorted
high = mid - 1; //the target would only be in lower part
else //the lower part is sorted
if(target < array[low])//the target is less than all elements in low part
low = mid + 1;
else
high = mid - 1;
else if(array[mid] < target)
if (array[low] < array[mid])// the lower part is sorted
low = mid + 1; //the target would only be in higher part
else //the higher part is sorted
if (array[high] < target)//the target is larger than all elements in higher part
high = mid - 1;
else
low = mid + 1;
else //if(array[mid] == target)
return mid;
}
return -1;
}
int search_ref(vector<int>& nums, int target) { int l = 0, r = nums.size() - 1; while (l <= r) { int mid = (l + r) / 2; if (target == nums[mid]) return mid; // there exists rotation; the middle element is in the left part of the array if (nums[mid] > nums[r]) { if (target < nums[mid] && target >= nums[l]) r = mid - 1; else l = mid + 1; } // there exists rotation; the middle element is in the right part of the array else if (nums[mid] < nums[l]) { if (target > nums[mid] && target <= nums[r]) l = mid + 1; else r = mid - 1; } // there is no rotation; just like normal binary search else { if (target < nums[mid]) r = mid - 1; else l = mid + 1; } } return -1; }
对比普通的二分查找法,为了确定目标数会落在二分后的那个部分,我们需要更多的判定条件。但是我们还是实现了O(log n)的目标。
二分查找法的缺陷
二分查找法的O(log n)让它成为十分高效的算法。不过它的缺陷却也是那么明显的。就在它的限定之上:
必须有序,我们很难保证我们的数组都是有序的。当然可以在构建数组的时候进行排序,可是又落到了第二个瓶颈上:它必须是数组。
数组读取效率是O(1),可是它的插入和删除某个元素的效率却是O(n)。因而导致构建有序数组变成低效的事情。
解决这些缺陷问题更好的方法应该是使用二叉查找树了,最好自然是自平衡二叉查找树了,自能高效的(O(n log n))构建有序元素集合,又能如同二分查找法一样快速(O(log n))的搜寻目标数。
#include <iostream> using namespace std; int binarysearch(int array[], int low, int high, int target) { if (low > high) return -1; int mid = (low + high) / 2; if (array[mid] > target) return binarysearch(array, low, mid - 1, target); if (array[mid] < target) return binarysearch(array, mid + 1, high, target); //if (midValue == target) return mid; } //Find the fisrt element, whose value is larger than target, in a sorted array int BSearchUpperBound(int array[], int low, int high, int target) { //Array is empty or target is larger than any every element in array if (low > high || target >= array[high]) return -1; int mid = (low + high) / 2; while (high > low) { if (array[mid] > target) high = mid; else low = mid + 1; mid = (low + high) / 2; } return mid; } //Find the fisrt element, whose value is larger than target, in a sorted array int BSearchUpperBound_ext(int array[], int low, int high, int target) { //Array is empty or target is larger than any every element in array if (low > high || target > array[high]) return -1; int mid = (low + high) / 2; while (high > low) { if (array[mid] >= target) high = mid; else low = mid + 1; mid = (low + high) / 2; } return mid; } //Find the last element, whose value is less than target, in a sorted array int BSearchLowerBound(int array[], int low, int high, int target) { //Array is empty or target is less than any every element in array if (high < low || target <= array[low]) return -1; int mid = (low + high + 1) / 2; //make mid lean to large side while (low < high) { if (array[mid] < target) low = mid; else high = mid - 1; mid = (low + high + 1) / 2; } return mid; } //Find the last element, whose value is less than target, in a sorted array int BSearchLowerBound_ext(int array[], int low, int high, int target) { //Array is empty or target is less than any every element in array if (high < low || target < array[low]) return -1; int mid = (low + high + 1) / 2; //make mid lean to large side while (low < high) { if (array[mid] <= target) low = mid; else high = mid - 1; mid = (low + high + 1) / 2; } return mid; } //return type: pair<int, int> //the fisrt value indicate the begining of range, //the second value indicate the end of range. //If target is not find, (-1,-1) will be returned pair<int, int> SearchRange(int A[], int n, int target) { pair<int, int> r(-1, -1); if (n <= 0) return r; int lower = BSearchLowerBound(A, 0, n - 1, target); lower = lower + 1; //move to next element if (A[lower] == target) r.first = lower; else //target is not in the array return r; int upper = BSearchUpperBound(A, 0, n - 1, target); upper = upper < 0 ? (n - 1) : (upper - 1); //move to previous element //since in previous search we had check whether the target is //in the array or not, we do not need to check it here again r.second = upper; return r; } int SearchInRotatedSortedArray(int array[], int low, int high, int target) { while (low <= high) { int mid = (low + high) / 2; if (target < array[mid]) if (array[mid] < array[high])//the higher part is sorted high = mid - 1; //the target would only be in lower part else //the lower part is sorted if (target < array[low])//the target is less than all elements in low part low = mid + 1; else high = mid - 1; else if (array[mid] < target) if (array[low] < array[mid])// the lower part is sorted low = mid + 1; //the target would only be in higher part else //the higher part is sorted if (array[high] < target)//the target is larger than all elements in higher part high = mid - 1; else low = mid + 1; else //if(array[mid] == target) return mid; } return -1; } int main() { int A[] = { 5, 7, 7, 8, 8, 10 }; int up=BSearchUpperBound(A,0,5, 8); //5 int low = BSearchLowerBound(A, 0, 5, 8);//2 auto range = SearchRange(A, 6, 8); //3,4 int up1 = BSearchLowerBound(A, 0, 5, 5); //-1 int low1 = BSearchUpperBound(A, 0, 5, 10); //-1 // test 非严格上下界 int up2 = BSearchLowerBound_ext(A, 0, 5, 8); //4 int low2 = BSearchUpperBound_ext(A, 0, 5, 8);//3 return 0; }
reference: