LeetCode "Top K Frequent Elements"

A typical solution is heap based - "top K". Complexity is O(nlgk).

typedef pair<int, unsigned> Rec;
struct Comp
{
    bool operator()(const Rec &r1, const Rec &r2)
    {
        return r1.second > r2.second;
    }
};
class Solution {

public:
    vector<int> topKFrequent(vector<int>& nums, int k) {
        unordered_map<int, unsigned> hm;
        for(auto&v : nums) hm[v]++;

        priority_queue<Rec, vector<Rec>, Comp> q;
        for(auto &kv : hm)
        {
            Rec r(kv.first, kv.second);
            q.push(r);
            if(q.size() > k)    q.pop();
        }

        vector<int> ret;
        while(!q.empty())
        {
            ret.push_back(q.top().first);
            q.pop();
        }
        return ret;
    }
};

There is a O(n) one indeed - bucketing the frequencies.
https://leetcode.com/discuss/100636/c-o-nlogk-and-o-n-solutions