LintCode 486: Merge K Sorted Arrays (经典题)

解法1：
利用priority queue + tuple。
记得我们需要最小堆，因为sorting是从小到大排序。而priority queue缺省是用最大堆，所以需要define class cmp 和 bool operator()。
tuple<i,j,k> 记录了k=arrays[i][j]。
时间复杂度应该是 O(N log k).
N is the total number of integers.
k is the number of arrays.

代码如下：

class cmp {
public:
    // to ensure a min-heap, otherwise it is max-heap by default
    bool operator() (tuple<int, int, int> &a, tuple<int, int, int> &b) {
        return get<2>(a) > get<2>(b);  
    }
};


class Solution {
public:
    /**
     * @param arrays: k sorted integer arrays
     * @return: a sorted array
     */
    vector<int> mergekSortedArrays(vector<vector<int>> &arrays) {
        
        tuple<int, int,  int>  tp;   //tuple<i,j,k> => k = arrays[i][j]
        priority_queue<tuple<int, int, int>, vector<tuple<int, int ,int>>, cmp> pq;
        vector<int> result;
        
        int arrayLen = arrays.size();
        for (int i  = 0; i < arrayLen; ++i) {
            if (arrays[i].size() != 0) {
                pq.push(make_tuple(i, 0, arrays[i][0]));
            }
        }
        
        while(!pq.empty()) {
            tuple<int, int, int> elem = pq.top();
            pq.pop();
            result.push_back(get<2>(elem));
            get<1>(elem)++;
            if (get<1>(elem) < arrays[get<0>(elem)].size()) {
                pq.push(make_tuple(get<0>(elem),
                                   get<1>(elem),
                                   arrays[get<0>(elem)][get<1>(elem)]));
            }
            
        }
        
        return result;
    }
};

解法2：就用priority_queue定义最小堆。
代码如下:

class Node {
public:
    Node(int r, int c, int v) : row(r), col(c), val(v) {}
    //min-heap
    bool operator < (const Node & obj) const {
        return val > obj.val;
    }
    int row, col, val;
};

class Solution {
public:
    /**
     * @param arrays: k sorted integer arrays
     * @return: a sorted array
     */
    vector<int> mergekSortedArrays(vector<vector<int>> &arrays) {
        vector<int> result;
        int nRow = arrays.size();
        if (nRow == 0) return result;
        int nCol = arrays[0].size();
        
        priority_queue<Node> pq;
        
        for (int i = 0; i < nRow; ++i) {
            if (arrays[i].size() > 0) pq.push(Node(i, 0, arrays[i][0]));
        }
        
        while(!pq.empty()) {
            Node currTop = pq.top();
            pq.pop();
            result.push_back(currTop.val);
            if (currTop.col + 1 < arrays[currTop.row].size()) {
                pq.push(Node(currTop.row, currTop.col + 1, arrays[currTop.row][currTop.col + 1]));
            }
        }
        
        return result;
    }
};

另外一种写法如下：

class Node {
public:
    Node(int r, int c, int v) : row(r), col(c), val(v) {}
    int row, col, val;
};

class cmp {
public:
    bool operator() (const Node & a, const Node & b) const {
        return a.val > b.val;
    }
};

class Solution {
public:
    /**
     * @param arrays: k sorted integer arrays
     * @return: a sorted array
     */
    vector<int> mergekSortedArrays(vector<vector<int>> &arrays) {
        vector<int> result;
        int nRow = arrays.size();
        if (nRow == 0) return result;
        int nCol = arrays[0].size();
        
        priority_queue<Node, vector<Node>, cmp> pq;
        
        for (int i = 0; i < nRow; ++i) {
            if (arrays[i].size() > 0) pq.push(Node(i, 0, arrays[i][0]));
        }
        
        while(!pq.empty()) {
            Node currTop = pq.top();
            pq.pop();
            result.push_back(currTop.val);
            if (currTop.col + 1 < arrays[currTop.row].size()) {
                pq.push(Node(currTop.row, currTop.col + 1, arrays[currTop.row][currTop.col + 1]));
            }
        }
        
        return result;
    }
};

3刷:
注意：这里是每行有序，所以用最小堆。每次
注意跟行列皆有序的LintCode 1874题的区分，那题也是用最小堆，但是没次先把最小的array[0][0]push到minHeap，然后每次pop出最小值，再把最小值的右边和下面元素push到minHeap。按那题的方法，minHeap的size()应该也是O(mnlogm)?
https://blog.csdn.net/roufoo/article/details/105308293

struct Node {
    int row, col, val;
    Node(int r, int c, int v) : row(r), col(c), val(v) {}
    bool operator < (const Node &node) const {
        return val > node.val;
    }
};
class Solution {
public:
    /**
     * @param arrays: k sorted integer arrays
     * @return: a sorted array
     */
    vector<int> mergekSortedArrays(vector<vector<int>> &arrays) {
        vector<int> res;
        int nRow = arrays.size();
        priority_queue<Node> minHeap;
        for (int i = 0; i < nRow; i++) {
            if (arrays[i].size() > 0) minHeap.push(Node(i, 0, arrays[i][0]));
        }
        while (!minHeap.empty()) {
            Node topNode = minHeap.top();
            minHeap.pop();
            res.push_back(topNode.val);
            if (topNode.col + 1 < arrays[topNode.row].size()) {
                minHeap.push(Node(topNode.row, topNode.col + 1, arrays[topNode.row][topNode.col + 1]));
            }
        }
        return res;
   }
};

解法3：还是minheap，但上面的时间复杂度是O(mnlogm)，下面的代码的解法的时间复杂度是O(mnlog(mn))，其实是把元素按每列的第1个元素，第2个元素这样按顺序压栈，再把top找出来，效率不高。上面的方法每次就把pop最小元素，同时把最小元素同行的下一个元素push进minHeap，这样minHeap的size一直都是k。而这里的方法每次把k个元素push进minHeap，但只pop出一个最小值，所以效率低。

class Solution {
public:
    /**
     * @param arrays: k sorted integer arrays
     * @return: a sorted array
     */
    vector<int> mergekSortedArrays(vector<vector<int>> &arrays) {
        vector<int> res;
        int nRow = arrays.size();
        vector<int> pointers(nRow, 0);
        priority_queue<int, vector<int>, greater<int>> minHeap;
        minHeap.push(INT_MIN);
        while (!minHeap.empty()) {
            for (int i = 0; i < nRow; i++) {
                if (pointers[i] >= arrays[i].size()) {
                    continue;
                };
                minHeap.push(arrays[i][pointers[i]]);
                pointers[i]++;
            }
            if (!minHeap.empty()) {
                res.push_back(minHeap.top());
                minHeap.pop();
            }
        }
        res.erase(res.begin()); //remove the fake INT_MIN pushed before the while() loop
        return res;        
    }
};