题目:
Given an array nums, there is a sliding window of size k which is moving from the very left of the array to the very right. You can only see the k numbers in the window. Each time the sliding window moves right by one position. Return the max sliding window.
Example:
Input: nums =[1,3,-1,-3,5,3,6,7]
, and k = 3 Output:[3,3,5,5,6,7] Explanation:
Window position Max --------------- ----- [1 3 -1] -3 5 3 6 7 3 1 [3 -1 -3] 5 3 6 7 3 1 3 [-1 -3 5] 3 6 7 5 1 3 -1 [-3 5 3] 6 7 5 1 3 -1 -3 [5 3 6] 7 6 1 3 -1 -3 5 [3 6 7] 7
Note:
You may assume k is always valid, 1 ≤ k ≤ input array's size for non-empty array.
Follow up:
Could you solve it in linear time?
最最简单的暴力brute force方法就是遍历整个数组的时候,对当前开始的size为k的window取最大。取最大可以简单粗暴地遍历一遍进行比较,这样是O(nk)的复杂度,但也可以直接用个heap来存然后取最top,也就是降到了O(nlogk)的复杂度。代码如下:
Runtime: 1700 ms, faster than 5.20% of C++ online submissions for Sliding Window Maximum.
Memory Usage: 179.7 MB, less than 6.56% of C++ online submissions for Sliding Window Maximum.
class Solution {
public:
vector<int> maxSlidingWindow(vector<int>& nums, int k) {
if (nums.empty()) {
return vector<int>();
}
vector<int> result;
for (int i = 0; i < nums.size() - k + 1; i++) {
priority_queue<int> window(nums.begin() + i, nums.begin() + i + k);
result.push_back(window.top());
}
return result;
}
};
另一种方法是采用deque,deque是双向队列,可以在任意一端进行push和pop。采用deque的思想在于,我们在deque中存放一段在window size范围内的非递增序列(也就是head元素是整个deque中最大的元素),每次遇到新的数字都把它push到deque里,并且如果这个数字比deque中的某些数字大,就把比它小的都pop掉,也就是相当于从tail开始把所有比当前数字小的都删掉,最后每次iteration的时候deque里的front就是目前window size中最大的元素。需要注意的就是要保证deque中的front一定要在window size范围内(对于这个实现方式的理解还需要更深刻一些)。这种方法的时间和空间复杂度都是O(n),其实这里刚开始有点不是特别理解时间复杂度为O(n),因为感觉在for循环里面也有个while,假如原数组是个递增的序列,那不是会while里面每次都运行k吗?其实并不会,因为在后面更大的数字被加入的时候,前面的就被remove了,所以其实相当于是数组中每个元素都被add和remove到deque里面一次,所以是O(n)。
https://www.zhihu.com/question/314669016/answer/663930108 这个链接中有很好的动画展示,真的非常易懂。
C++:
Runtime: 56 ms, faster than 86.96% of C++ online submissions for Sliding Window Maximum.
Memory Usage: 13.1 MB, less than 77.05% of C++ online submissions for Sliding Window Maximum.
class Solution {
public:
vector<int> maxSlidingWindow(vector<int>& nums, int k) {
if (nums.empty()) {
return vector<int>();
}
vector<int> result;
deque<int> dq;
for (int i = 0; i < nums.size(); i++) {
// should use dq.back() because it might be smaller than first
// but larger than remaining
while (!dq.empty() && nums[i] > dq.back()) {
dq.pop_back();
}
dq.push_back(nums[i]);
if (i >= k - 1) {
result.push_back(dq.front());
// if the current front is the first elem in the window
// we cannot keep it in the next iteration, so pop it
if (nums[i - k + 1] == dq.front()) {
dq.pop_front();
}
}
}
return result;
}
};
Java:
Runtime: 22 ms, faster than 34.04% of Java online submissions for Sliding Window Maximum.
Memory Usage: 60.4 MB, less than 28.40% of Java online submissions for Sliding Window Maximum.
class Solution {
public int[] maxSlidingWindow(int[] nums, int k) {
int[] result = new int[nums.length - k + 1];
Deque<Integer> deque = new ArrayDeque<>();
for (int i = 0; i < nums.length; i++) {
// pop the elements that are not larger than current element
// start from the tail of the deque
while (!deque.isEmpty() && nums[i] > deque.getLast()) {
deque.removeLast();
}
deque.addLast(nums[i]);
if (i >= k - 1) {
// store the head of the deque as the result
result[i - k + 1] = deque.getFirst();
// remove the head if it will not be in the window the next time
if (nums[i - k + 1] == deque.getFirst()) {
deque.removeFirst();
}
}
}
return result;
}
}
最后一种方法是有一点动态规划的思想。我们先把整个数组分成大小为k的几组,最后一组的个数可以小于k。然后我们分别对每组中的元素(注意是同一组)求出从左到右的max和从右到左的max(有那么一点点像trapping rain water)。我们在slide整个数组的时候,这个window可能有两种情况,一种是所有元素都在刚刚分出来的同一个组中,另一种是分别在两个不同的组中:
假如当前的窗口是从i到j,并且i是某一组的起始元素,那么当前window的max就是left[j]或者right[i]
假如i不是起始元素,那么一定是第二种情况,那么当前window的max应该是max(left[j], right[i]),因为left[j]是j所在的这组元素从左往右最大值,right[i]是i所在的这组元素从右往左的最大值,二者之中的最大一定就是当前window被分开的两边的最大值的其中之一。
两种情况可以合并,其实就直接是max(left[j], right[i])。这种做法的时空复杂度也都是O(n)。
Runtime: 56 ms, faster than 86.96% of C++ online submissions for Sliding Window Maximum.
Memory Usage: 14.3 MB, less than 22.95% of C++ online submissions for Sliding Window Maximum.
class Solution {
public:
vector<int> maxSlidingWindow(vector<int>& nums, int k) {
if (nums.empty()) {
return vector<int>();
}
vector<int> result, left, right;
int max_left, max_right;
for (int i = 0; i < nums.size(); i++) {
if (i % k == 0) {
max_left = nums[i];
}
else {
max_left = max(max_left, nums[i]);
}
left.push_back(max_left);
}
for (int i = nums.size() - 1; i >= 0; i--) {
if (i % k == 0) {
max_right = nums[i];
}
else {
max_right = max(max_right, nums[i]);
}
right.push_back(max_right);
}
reverse(right.begin(), right.end());
for (int i = 0; i < nums.size() - k + 1; i++) {
result.push_back(max(left[i + k - 1], right[i]));
}
return result;
}
};
上面代码中进行了两次for循环得到left和right数组,其实可以放在一个循环里,但是懒得写了……
然后补java的时候就把两次for循环合一起了,结果各种下标搞死我了,但主要感觉还是对这个做法的理解不够深入所以才会很晕,各种试和抄答案。
Runtime: 7 ms, faster than 90.96% of Java online submissions for Sliding Window Maximum.
Memory Usage: 60.1 MB, less than 31.52% of Java online submissions for Sliding Window Maximum.
class Solution {
public int[] maxSlidingWindow(int[] nums, int k) {
int[] result = new int[nums.length - k + 1];
int[] left = new int[nums.length];
int[] right = new int[nums.length];
for (int i = 0; i < nums.length; i++) {
if (i % k == 0) {
left[i] = nums[i];
} else {
left[i] = Math.max(left[i - 1], nums[i]);
}
int r = nums.length - i - 1;
// the start of r is the last element of a block, or, it is the last element
if ((r + 1) % k == 0 || r == nums.length - 1) {
right[r] = nums[r];
} else {
right[r] = Math.max(right[r + 1], nums[r]);
}
}
// the result is the max of the left[rightElement] and right[leftElement]
for (int i = 0; i < nums.length - k + 1; i++) {
result[i] = Math.max(left[i + k - 1], right[i]);
}
return result;
}
}