Given an N * M matrix mat[][], the task is to find the area-wise largest rectangular sub-matrix such that each column and each row of the sub-matrix is strictly increasing.
Examples:
Input: mat[][] =
{{1, 2, 3},
{4, 5, 6},
{1, 2, 3}}
Output: 6
Largest sub-matrix will be {{1, 2, 3}, {4, 5, 6}}.
Number of elements in this sub-matrix = 6.Input: mat[][] =
{{1, 2, 3},
{4, 5, 3},
{1, 2, 3}}
Output: 4
The largest sub-matrix will be
{{1, 2}, {4, 5}}
Approach: There are many approaches to solve this problem varying from O(N3 * M3) to O(N * M). In this article, an approach with O(N * M) time complexity using a stack will be discussed.
Before proceeding further, its recommended to solve this. problem.
Let’s try to understand the approach broadly, then the algorithm will be discussed. For every column of the matrix, try to find the largest row-wise and column-wise sorted sub-matrix having the left edge at this column. To perform the same, create a matrix pre[][] where pre[i][j] will store the length of the longest increasing sub-array starting from the index j of the array arr[i].
Now using this matrix, for each column j, find the length of the longest row-wise and column-wise sorted array. To process a column, all the increasing sub-segments of the array pre[][j] will be required. The same can be found using the two-pointer technique. In each of these sub-segments, simply find the largest area under the histogram considering the row-wise increasing sub-segments as bars.
- Create a prefix-sum array for each row ‘i’, which stores length of the largest increasing sub-array ending at each column ‘j’ of that row.
- Once we have this array, for each column ‘j’.
- Initialize ‘i’ equals 0.
- Run a loop on ‘i’ while ‘i’ is less than ‘N’
- Initialize ‘k’ equals i+1.
- while k less than N and arr[k][j] greater than arr[k-1][j], increment k.
- Apply histogram problem on the sub-array pre[i][j] to pre[k-1][j], to find the largest area under it. Let us call this value ‘V’. Update final answer as ans = max(ans, val).
- Update ‘i’ equals k-1.
Below is the implementation of the above approach:
C++
// C++ implementation of the approach #include <bits/stdc++.h> using namespace std; // Function to return the largest // area under a histogram int histo(vector< int > q) { // Stack stack< int > q1; // Length of the vector int n = q.size(); // Function to store the next smaller // and previous smaller index int pre_smaller[q.size()]; int next_smaller[q.size()]; // Finding the next smaller for ( int i = 0; i < n; i++) pre_smaller[i] = -1, next_smaller[i] = n; for ( int i = 0; i < n; i++) { while (q1.size() and q[q1.top()] > q[i]) { next_smaller[q1.top()] = i; q1.pop(); } q1.push(i); } // Finding the previous smaller element while (q1.size()) q1.pop(); for ( int i = n - 1; i >= 0; i--) { while (q1.size() and q[q1.top()] > q[i]) { pre_smaller[q1.top()] = i; q1.pop(); } q1.push(i); } // To store the final answer int ans = 0; // Finding the final answer for ( int i = 0; i < n; i++) ans = max(ans, (next_smaller[i] - pre_smaller[i] - 1) * q[i]); // Returning the final answer return ans; } // Function to return the largest area // for the required submatrix int findLargest(vector<vector< int > > arr) { // n and m store the number of // rows and columns respectively int n = arr.size(); int m = arr[0].size(); // To store the prefix_sum int pre[n][m]; // To store the final answer int ans = 0; // Loop to create the prefix-sum // using two pointers for ( int i = 0; i < n; i++) for ( int j = 0; j < m; j++) { if (j == 0) { pre[i][j] = 1; continue ; } if (arr[i][j] > arr[i][j - 1]) pre[i][j] = pre[i][j - 1] + 1; else pre[i][j] = 1; } // For each column run the loop for ( int j = 0; j < m; j++) { // Find the largest row-wise sorted arrays for ( int i = 0; i < n; i++) { int k = i + 1; vector< int > q; q.push_back(pre[i][j]); while (k < n and arr[k] > arr[k - 1]) q.push_back(pre[k][j]), k++; // Applying the largest area // under the histogram ans = max(ans, histo(q)); i = k - 1; } } // Return the final answer return ans; } // Driver code int main() { vector<vector< int > > arr = { { 1, 2, 3 }, { 4, 5, 6 }, { 1, 2, 3 } }; cout << findLargest(arr); return 0; } |
Java
// Java implementation of the approach import java.io.*; import java.util.*; class GFG { // Function to return the largest // area under a histogram static int histo(ArrayList<Integer> q) { // Stack Stack<Integer> q1 = new Stack<Integer>(); // Length of the vector int n = q.size(); // Function to store the next smaller // and previous smaller index int [] pre_smaller = new int [q.size()]; int [] next_smaller = new int [q.size()]; // Finding the next smaller for ( int i = 0 ; i < n; i++) { pre_smaller[i] = - 1 ; next_smaller[i] = n; } for ( int i = 0 ; i < n; i++) { while (q1.size() > 0 && q.get(q1.peek()) > q.get(i)) { next_smaller[q1.peek()] = i; q1.pop(); } q1.push(i); } // Finding the previous smaller element while (q1.size() > 0 ) { q1.pop(); } for ( int i = n - 1 ; i >= 0 ; i--) { while (q1.size() > 0 && q.get(q1.peek()) > q.get(i)) { pre_smaller[q1.peek()] = i; q1.pop(); } q1.push(i); } // To store the final answer int ans = 0 ; // Finding the final answer for ( int i = 0 ; i < n; i++) { ans = Math.max(ans, (next_smaller[i] - pre_smaller[i] - 1 ) * q.get(i)); } // Returning the final answer return ans; } // Function to return the largest area // for the required submatrix static int findLargest(ArrayList<ArrayList<Integer>> arr) { // n and m store the number of // rows and columns respectively int n = arr.size(); int m = arr.get( 0 ).size(); // To store the prefix_sum int [][] pre= new int [n][m]; // To store the final answer int ans = 0 ; // Loop to create the prefix-sum // using two pointers for ( int i = 0 ; i < n; i++) { for ( int j = 0 ; j < m; j++) { if (j == 0 ) { pre[i][j] = 1 ; continue ; } if (arr.get(i).get(j) > arr.get(i).get(j - 1 )) { pre[i][j] = pre[i][j - 1 ] + 1 ; } else { pre[i][j] = 1 ; } } } // For each column run the loop for ( int j = 0 ; j < m; j++) { // Find the largest row-wise sorted arrays for ( int i = 0 ; i < n; i++) { int k = i + 1 ; ArrayList<Integer> q = new ArrayList<Integer>(); q.add(pre[i][j]); while (k < n && arr.get(k).get( 0 ) > arr.get(k - 1 ).get( 0 )) { q.add(pre[k][j]); k++; } // Applying the largest area // under the histogram ans = Math.max(ans, histo(q)); i = k - 1 ; } } // Return the final answer return ans; } // Driver code public static void main (String[] args) { ArrayList<ArrayList<Integer>> arr = new ArrayList<ArrayList<Integer>>(); arr.add( new ArrayList<Integer>(Arrays.asList( 1 , 2 , 3 ))); arr.add( new ArrayList<Integer>(Arrays.asList( 4 , 5 , 6 ))); arr.add( new ArrayList<Integer>(Arrays.asList( 1 , 2 , 3 ))); System.out.println(findLargest(arr)); } } // This code is contributed by avanitrachhadiya2155 |
Python3
# Python3 implementation of the approach # Function to return the largest # area under a histogram def histo(q): # Stack q1 = [] # Length of the vector n = len (q) # Function to store the next smaller # and previous smaller index pre_smaller = [ 0 for i in range ( len (q))] next_smaller = [ 0 for i in range ( len (q))] # Finding the next smaller for i in range (n): pre_smaller[i] = - 1 next_smaller[i] = n for i in range (n): while ( len (q1) > 0 and q[q1[ - 1 ]] > q[i]): next_smaller[q1[ - 1 ]] = i del q1[ - 1 ] q1.append(i) # Finding the previous smaller element while ( len (q1) > 0 ): del q1[ - 1 ] for i in range (n - 1 , - 1 , - 1 ): while ( len (q1) > 0 and q[q1[ - 1 ]] > q[i]): pre_smaller[q1[ - 1 ]] = i del q1[ - 1 ] q1.append(i) # To store the final answer ans = 0 # Finding the final answer for i in range (n): ans = max (ans, (next_smaller[i] - pre_smaller[i] - 1 ) * q[i]) # Returning the final answer return ans # Function to return the largest area # for the required submatrix def findLargest(arr): # n and m store the number of # rows and columns respectively n = len (arr) m = len (arr[ 0 ]) # To store the prefix_sum pre = [[ 0 for i in range (m)] for i in range (n)] # To store the final answer ans = 0 # Loop to create the prefix-sum # using two pointers for i in range (n): for j in range (m): if (j = = 0 ): pre[i][j] = 1 continue if (arr[i][j] > arr[i][j - 1 ]): pre[i][j] = pre[i][j - 1 ] + 1 else : pre[i][j] = 1 # For each column run the loop for j in range (m): # Find the largest row-wise sorted arrays for i in range (n): k = i + 1 q = [] q.append(pre[i][j]) while (k < n and arr[k] > arr[k - 1 ]): q.append(pre[k][j]) k + = 1 # Applying the largest area # under the histogram ans = max (ans, histo(q)) i = k - 1 # Return the final answer return ans # Driver code arr = [ [ 1 , 2 , 3 ], [ 4 , 5 , 6 ], [ 1 , 2 , 3 ] ] print (findLargest(arr)) # This code is contributed by mohit kumar 29 |
C#
// C# implementation of the approach using System; using System.Collections.Generic; class GFG{ // Function to return the largest // area under a histogram static int histo(List< int > q) { // Stack Stack< int > q1 = new Stack< int >(); // Length of the vector int n = q.Count; // Function to store the next smaller // and previous smaller index int [] pre_smaller = new int [q.Count]; int [] next_smaller = new int [q.Count]; // Finding the next smaller for ( int i = 0; i < n; i++) { pre_smaller[i] = -1; next_smaller[i] = n; } for ( int i = 0; i < n; i++) { while (q1.Count > 0 && q[q1.Peek()] > q[i]) { next_smaller[q1.Peek()] = i; q1.Pop(); } q1.Push(i); } // Finding the previous smaller element while (q1.Count > 0) { q1.Pop(); } for ( int i = n - 1; i >= 0; i--) { while (q1.Count > 0 && q[q1.Peek()] > q[i]) { pre_smaller[q1.Peek()] = i; q1.Pop(); } q1.Push(i); } // To store the final answer int ans = 0; // Finding the final answer for ( int i = 0; i < n; i++) { ans = Math.Max(ans, (next_smaller[i] - pre_smaller[i] - 1) * q[i]); } // Returning the // final answer return ans; } // Function to return the largest area // for the required submatrix static int findLargest(List<List< int >> arr) { // n and m store the number of // rows and columns respectively int n = arr.Count; int m = arr[0].Count; // To store the prefix_sum int [,] pre = new int [n, m]; // To store the final answer int ans = 0; // Loop to create the prefix-sum // using two pointers for ( int i = 0; i < n; i++) { for ( int j = 0; j < m; j++) { if (j == 0) { pre[i, j] = 1; continue ; } if (arr[i][j] > arr[i][j - 1]) { pre[i, j] = pre[i,j - 1] + 1; } else { pre[i, j] = 1; } } } // For each column run the loop for ( int j = 0; j < m; j++) { // Find the largest row-wise sorted arrays for ( int i = 0; i < n; i++) { int k = i + 1; List< int > q = new List< int >(); q.Add(pre[i, j]); while (k < n && arr[k][0] > arr[k - 1][0]) { q.Add(pre[k, j]); k++; } // Applying the largest area // under the histogram ans = Math.Max(ans, histo(q)); i = k - 1; } } // Return the final answer return ans; } // Driver code static public void Main() { List<List< int >> arr = new List<List< int >>(); arr.Add( new List< int >(){1, 2, 3}); arr.Add( new List< int >(){4, 5, 6 }); arr.Add( new List< int >(){1, 2, 3}); Console.WriteLine(findLargest(arr)); } } // This code is contributed by rag2127 |
Javascript
<script> // Javascript implementation of the approach // Function to return the largest // area under a histogram function histo(q) { // Stack let q1 = []; // Length of the vector let n = q.length; // Function to store the next smaller // and previous smaller index let pre_smaller = new Array(q.length); let next_smaller = new Array(q.length); // Finding the next smaller for (let i = 0; i < n; i++) { pre_smaller[i] = -1; next_smaller[i] = n; } for (let i = 0; i < n; i++) { while (q1.length > 0 && q[q1[q1.length-1]] > q[i]) { next_smaller[q1[q1.length-1]] = i; q1.pop(); } q1.push(i); } // Finding the previous smaller element while (q1.length > 0) { q1.pop(); } for (let i = n - 1; i >= 0; i--) { while (q1.length > 0 && q[q1[q1.length-1]] > q[i]) { pre_smaller[q1[q1.length-1]] = i; q1.pop(); } q1.push(i); } // To store the final answer let ans = 0; // Finding the final answer for (let i = 0; i < n; i++) { ans = Math.max(ans, (next_smaller[i] - pre_smaller[i] - 1) * q[i]); } // Returning the final answer return ans; } // Function to return the largest area // for the required submatrix function findLargest(arr) { // n and m store the number of // rows and columns respectively let n = arr.length; let m = arr[0].length; // To store the prefix_sum let pre= new Array(n); for (let i=0;i<n;i++) { pre[i]= new Array(m); for (let j=0;j<m;j++) { pre[i][j]=0; } } // To store the final answer let ans = 0; // Loop to create the prefix-sum // using two pointers for (let i = 0; i < n; i++) { for (let j = 0; j < m; j++) { if (j == 0) { pre[i][j] = 1; continue ; } if (arr[i][j] > arr[i][j - 1]) { pre[i][j] = pre[i][j - 1] + 1; } else { pre[i][j] = 1; } } } // For each column run the loop for (let j = 0; j < m; j++) { // Find the largest row-wise sorted arrays for (let i = 0; i < n; i++) { let k = i + 1; let q = []; q.push(pre[i][j]); while (k < n && arr[k][0] > arr[k - 1][0]) { q.push(pre[k][j]); k++; } // Applying the largest area // under the histogram ans = Math.max(ans, histo(q)); i = k - 1; } } // Return the final answer return ans; } // Driver code let arr=[[1, 2, 3],[4, 5, 6 ],[1, 2, 3]]; document.write(findLargest(arr)); // This code is contributed by patel2127 </script> |
6
Time Complexity: O(N*N), as we are using nested loops to traverse N*N times.
Auxiliary Space: O(N*N), as we are using extra space for matrix.
Ready to dive in? Explore our Free Demo Content and join our DSA course, trusted by over 100,000 neveropen!