716. Max Stack

Design a max stack that supports push, pop, top, peekMax and popMax.

push(x) – Push element x onto stack.
pop() – Remove the element on top of the stack and return it.
top() – Get the element on the top.
peekMax() – Retrieve the maximum element in the stack.
popMax() – Retrieve the maximum element in the stack, and remove it. If you find more than one maximum elements, only remove the top-most one.
Example 1:
MaxStack stack = new MaxStack();
stack.push(5);
stack.push(1);
stack.push(5);
stack.top(); -> 5
stack.popMax(); -> 5
stack.top(); -> 1
stack.peekMax(); -> 5
stack.pop(); -> 1
stack.top(); -> 5
Note:

1. -1e7 <= x <= 1e7
2. Number of operations won’t exceed 10000.
3. The last four operations won’t be called when stack is empty.

方法1: two stacks

思路：

参照155. Min Stacks，再加一个popMax的功能。popMax时用一个buffer来缓存从st中寻找最大值时暂时倒出来的数字，同时helper对应数字也被pop出来。找到后，两个栈同时pop掉最大值，调用push将buffer中的数字再存回去。注意清空和放回要用MaxStack的api才能保证helper被同时更新。

Complexity

Time complexity: 1 ~ 4, O(1), 5, O(n)
Space complexity: O(n)

class MaxStack {
private:
    stack<int> st, helper;
public:
    /** initialize your data structure here. */
    MaxStack() {
        
    }
    
    void push(int x) {
        if (!helper.empty()) {
            x > helper.top() ? helper.push(x) : helper.push(helper.top());
        } else {
            helper.push(x);
        }
        st.push(x);
    }
    
    int pop() {
        int val = st.top();
        st.pop();
        helper.pop();
        return val;
    }
    
    int top() {
        return st.top();
    }
    
    int peekMax() {
        return helper.top();
    }
    
    int popMax() {
        int val = helper.top();
        stack<int> buffer;
        while (st.top() != val) {
            buffer.push(st.top());
            st.pop();
            helper.pop();
        }
        st.pop();
        helper.pop();
        while (!buffer.empty()) {
            push(buffer.top());
            buffer.pop();
        }
        return val;
    }
};

/**
 * Your MaxStack object will be instantiated and called as such:
 * MaxStack* obj = new MaxStack();
 * obj->push(x);
 * int param_2 = obj->pop();
 * int param_3 = obj->top();
 * int param_4 = obj->peekMax();
 * int param_5 = obj->popMax();
 */

方法2: doubly linked list + treemap

官方题解：https://leetcode.com/problems/max-stack/solution/

思路：

为了使popMax的时间降到O(n)以下，用doubly linked list 来保存数字，类似于之前的stack，但是可以做到O(1) 时间删除某一个节点。但是要怎么知道最大值保存在哪里呢？这就要借助一个辅助map来保存每一个val所对应在dll中的所有iterator，用val作为key，用一个vector来作为hash的val。而为了随时可以找到val中的最大值，需要用treemap来对这些key进行排序。这样在删除时（pop, poMax），可以找到要删除的val，然后在其对应的vector 中找到最后一个值，用这个iterator在dll中用O(1)完成删除，再将iterator本身从hash中pop掉。还要注意如果hash[val]此时为空的话， 需要将val从hash中删除，不要误导下面的getMax。

Complexity

Time complexity: O(log n)
Space complexity: O(n)

class MaxStack {
private:
    list<int> dll;
    map<int, vector<list<int>::iterator>> hash;
public:
    /** initialize your data structure here. */
    MaxStack() {
        
    }
    
    void push(int x) {
        dll.push_front(x);
        hash[x].push_back(dll.begin());
    }
    
    int pop() {
        int val = dll.front();
        hash[val].pop_back();
        dll.pop_front();
        if (hash[val].empty()) hash.erase(val);
        return val;
    }
    
    int top() {
        return dll.front();
    }
    
    int peekMax() {
        return hash.rbegin()->first;
    }
    
    int popMax() {
        int val = hash.rbegin()->first;
        auto it = hash[val].back();
        dll.erase(it);
        hash[val].pop_back();
        if (hash[val].empty()) hash.erase(val);
        return val;
    }
};

/**
 * Your MaxStack object will be instantiated and called as such:
 * MaxStack* obj = new MaxStack();
 * obj->push(x);
 * int param_2 = obj->pop();
 * int param_3 = obj->top();
 * int param_4 = obj->peekMax();
 * int param_5 = obj->popMax();
 */