算法题目-记hulu失败的实习面试
1。对于数组A[0,1,2,3,4,...,k],求得0<=i < j < k,且使得A[j] - A[i]为最大值。
最简单也最容易想到的搜索两遍,即可得到答案。i的位置从起始至倒数第二个位置,j的位置从末尾元素至i后一个位置,保存记录最大的差值即可。
不过最简单的方法复杂度为n的平方,其实令有一个时间复杂度很低的方法,及从前至后遍历,添加一个保存当前访问元素之前的最小的元素,最大值必定需要减去已访问过元素的最小值才能够获得,这样时间复杂度降至n。
class Solution {
public: //i<j max(a[j]-a[i]) int arrayMaxGap(vector<int> &array) { if(array.size() == 0 || array.size() == 1) return -1; if(array.size() == 2) { return array[1] - array[0]; } int maxres,premin; maxres = array[1] - array[0]; premin = array[0]; for(int i = 1;i < array.size();i++) { if (array[i] - premin > maxres) maxres = array[i] - premin; if (array[i] < premin) premin = array[i]; } return maxres; } }
2。对于递增的数组A[0,1,2,3,4,...,k],数组B[0,1,2,3,4,...k'],对于0<=i<k,0<=j<k',对于计算出的A[i]+B[j],求其前k小个元素。
对于这个题目最先想到的思路,构造一个集合计算出所有的A[i]+B[j],然后求取前k小个元素就比较简单了,但是如果k相对于构造出的集合比较小,这样就比较浪费空间了。
对于任意一个元素A[i]+B[j],我们可以指导A[i+1]+B[j]和A[i]+B[j+1]是其接下来可确定的最近邻的两个大于其的元素,首先A[0]+B[0],接下来就是A[1]+B[0],A[0]+B[1]之中小的元素为第二小的元素,接下来继续扩展第二小的元素得到A[2]+B[0],A[1]+B[1],或者A[0]+B[2],A[1]+B[1],所以如果用指针的话这样无法保存,求取最小的结合不断的膨胀,所以这里考虑利用最小堆每次得到多个元素中最小的元素。
//class for minKsum struct Point { int x; int y; int sum; }; bool operator<(Point a,Point b){ return a.sum > b.sum; } class Solution { public: //array A,arrayB, obtain the minest k a[i]+b[j]. that a[0]+b[0] is the minest vector<int> minKsum(vector<int> i_A,vector<int> i_B,int k) { vector<int> res; priority_queue<Point> priorityq; if(i_A.size() == 0&&i_B.size() == 0) { return res; } if(i_A.size() == 0) { if (i_B.size() > k) { return res; } return vector<int>(i_B.begin(),i_B.begin()+k); } if(i_B.size() == 0) { if(i_A.size() > k) { return res; } return vector<int>(i_A.begin(),i_A.begin()+k); } Point first; first.x = 0; first.y = 0; first.sum = i_A[0] + i_B[0]; priorityq.push(first); while (!priorityq.empty()&&res.size() != k) { Point tmpoutput = priorityq.top(); priorityq.pop(); res.push_back(tmpoutput.sum); if(tmpoutput.y == 0) { if (tmpoutput.x + 1 < i_A.size()) { Point tmp; tmp.x = tmpoutput.x + 1; tmp.y = tmpoutput.y; tmp.sum = i_A[tmpoutput.x + 1] + i_B[tmpoutput.y]; priorityq.push(tmp); } if (tmpoutput.y + 1 < i_B.size()) { Point tmp; tmp.x = tmpoutput.x; tmp.y = tmpoutput.y + 1; tmp.sum = i_A[tmpoutput.x] + i_B[tmpoutput.y + 1]; priorityq.push(tmp); } } else { if(tmpoutput.y + 1 < i_B.size()) { Point tmp; tmp.x = tmpoutput.x; tmp.y = tmpoutput.y + 1; tmp.sum = i_A[tmpoutput.x] + i_B[tmpoutput.y + 1]; priorityq.push(tmp); } } } return res; } }
3。实现带有重复元素的二分查找,如果查找的元素重复,返回重复元素的起始位置。
类似二分查找,但是结束条件有些不同,结束条件为当前节点=目标元素并且当前节点为起始节点或者前一个节点不等与目标元素。
class Solution {
public: int binarySearch(vector<int> &array, int target) { if(array.size() == 1) { if (array[0] == target) return 0; else return -1; } int start = 0,end = array.size() - 1,mid; while(start <= end) { mid = (start + end)/2; if(array[mid] == target && (mid == 0 || array[mid-1] != target)) { return mid; } else if(target <= array[mid]) { end = mid - 1; } else { start = mid + 1; } } return -1; } }
4。对于链表表示的完全二叉树,给出根节点的指针,统计其节点的数量。
完全遍历需要O(n)时间的复杂度。但是这样没有完全利用完全二叉树的特性。
计算节点数量转换为计算最后一层节点数量,计算最后一层节点数量可以这样计算,左子树,右子树,如果深度相同,计算右子树最后一层数量+满的左子树。如果深度不同,计算左子树的最后一层节点数量,这样问题规模可以直接减半。
#include <algorithm> #include <time.h> #include <iostream> using namespace std; struct TreeNode { TreeNode *left; TreeNode *right; int value; TreeNode(int v) {value = v;} }; class Solution { public: int recurseTree(TreeNode *root) { if (root == NULL) return 0; int tdepth = depth(root); return (int)pow((double)2,tdepth-1) - 1 + recuseNode(root,tdepth); } //递归统计完全二叉树的最后一层节点个数 int recuseNode(TreeNode *root,int depth) { if(root == NULL) return 0; //最后一层才进行统计,depth == 1 if(root->left == NULL && root->right == NULL && depth == 1) return 1; return recuseNode(root->left,depth-1)+recuseNode(root->right,depth-1); } int dc(TreeNode *root) { if(root == NULL) return 0; int treedepth = depth(root); if(treedepth == 1) return 1; else return (int)pow((double)2,treedepth-1) - 1 + devide_conquer(root); } //分治统计完全二叉树的最后一层节点个数 int devide_conquer(TreeNode *root) { int tdepth = depth(root); //如果只有两层,直接可以得到最后一层节点个数 if(tdepth == 2) { if(root->right == NULL) return 1; return 2; } int leftdepth = depth(root->left); int rightdepth = depth(root->right); //right side if (leftdepth == rightdepth) { return (int)pow((double)2,tdepth-1-1) + devide_conquer(root->right); } //left side else if(leftdepth > rightdepth) { return devide_conquer(root->left); } else { //error } } //计算深度 int depth(TreeNode *root) { int depth = 0; TreeNode *cur = root; while(cur != NULL) { cur = cur->left; depth++; } return depth; } }; int main(int argc,char **argv) { TreeNode *_1root = new TreeNode(1); _1root->left = NULL; _1root->right = NULL; TreeNode *_2root1 = new TreeNode(1); TreeNode *_2root2 = new TreeNode(2); TreeNode *_2root3 = new TreeNode(3); TreeNode *_2root4 = new TreeNode(4); TreeNode *_2root5 = new TreeNode(5); TreeNode *_2root6 = new TreeNode(6); TreeNode *_2root7 = new TreeNode(7); TreeNode *_2root8 = new TreeNode(8); TreeNode *_2root9 = new TreeNode(9); TreeNode *_2root10 = new TreeNode(10); TreeNode *_2root11 = new TreeNode(11); TreeNode *_2root12 = new TreeNode(12); _2root1->left = _2root2; _2root1->right = _2root3; _2root2->left = _2root4; _2root2->right = _2root5; _2root3->left = _2root6; _2root3->right = _2root7; _2root4->left = _2root8; _2root4->right = _2root9; _2root5->left = _2root10; _2root5->right = _2root11; _2root6->left = _2root12; _2root6->right = NULL; _2root7->left = NULL; _2root7->right = NULL; _2root8->left = NULL; _2root8->right = NULL; _2root9->left = NULL; _2root9->right = NULL; _2root10->left = NULL; _2root10->right = NULL; _2root11->left = NULL; _2root11->right = NULL; _2root12->left = NULL; _2root12->right = NULL; Solution s; int s11,s12,s21,s22; clock_t start,finish; start = clock(); for(int i = 0;i < 1000000;i++) s11 = s.dc(_1root); finish = clock(); double t11 = double(finish - start); start = clock(); for(int i = 0;i < 1000000;i++) s12 = s.recurseTree(_1root); finish = clock(); double t12 = double(finish - start); start = clock(); for(int i = 0;i < 1000000;i++) s21 = s.dc(_2root1); finish = clock(); double t21 = double(finish - start); start = clock(); for(int i = 0;i < 1000000;i++) s22 = s.recurseTree(_2root1); finish = clock(); double t22 = double(finish - start); cout<<"s11:"<<s11<<"time:"<<t11<<"ms"<<endl; cout<<"s12:"<<s12<<"time:"<<t12<<"ms"<<endl; cout<<"s21:"<<s21<<"time:"<<t21<<"ms"<<endl; cout<<"s22:"<<s22<<"time:"<<t22<<"ms"<<endl; }
5。给出一串数字,判断是否为有效二叉查找树的后序遍历序列(及是否能够通过这串后序遍历序列构造出二叉查找树)。
最先想到的思路,将当前序列排序,有序的假设为中序遍历序列,这样可以类似还原BST。利用后序的序列不断的分割中序。有序的序列是否能够在后序中为一串连续的空间,如果连续即可构造出BST。这种方法的时间复杂度可以从O(n2)优化至O(nlgn),利用hash表。
有一个更好的思路,不利用中序,直接利用后序,首先利用最后的根节点,利用二分查找nlgn得到分割点,当前元素小于根,下一个元素大于根,考虑一些边界情况。
然后,根节点之前的为左子树,其上界为根节点;根节点之后的为右子树,其下界为根节点,递归下去,递归至一个元素,如果符合上下界的条件,即可形成BST。
class Solution {
public: //postorder sequece and checkout if it can be construct BST bool validBST(vector<int> &postorder) { return validBST_recurse(postorder, 0, postorder.size() - 1, INT_MIN, INT_MAX); } bool validBST_recurse(vector<int> &postorder, int start, int end, int min, int max) { if(end < start) { return true; } if(end == start) { return postorder[end] > min && postorder[end] < max; } int partition = binaryPartition(postorder,start,end); bool left = validBST_recurse(postorder, start, partition, min, postorder[end]); bool right = validBST_recurse(postorder, partition+1, end-1, postorder[end], max); return left&&right; } //partition to two part, postorder[p] < postorder[end] && portorder[p+1] > postorder[end] int binaryPartition(vector<int> &postorder, int start, int end) { int target = postorder[end]; int mid; end = end - 1; if(start == end) return start; while(start < end) { mid = (start + end)/2; if(postorder[mid] < target && (mid == end || postorder[mid+1] > target)) { return mid; } else if(postorder[mid] > target && (mid == start || postorder[mid-1] < target)) { return mid-1; } else if(postorder[mid] < target) { start = mid + 1; } else { end = mid - 1; } } return -1; } }