Median of sliding window in an array

Given an array of integer arr[] and an integer k, the task is to find the median of each window of size k starting from the left and moving towards the right by one position each time.

Examples:

Input: arr[] = {-1, 5, 13, 8, 2, 3, 3, 1}, k = 3
Output: 5 8 8 3 3 3



Input: arr[] = {-1, 5, 13, 8, 2, 3, 3, 1}, k = 4
Output: 6.5 6.5 5.5 3.0 2.5

Approach: Create a pair class to hold the items and their index. It also implements the comparable interface so that compareTo() method will be invoked by the Treeset to find the nodes. Note that the two pairs are equal only when their indices are equal. This is important since a window can contain duplicates and we may end up deleting multiple items in single remove() call if we only check for the value.

The idea is to maintain two sorted sets (minSet and maxSet) of Pair objects of length (k / 2) and (k / 2) + 1 depending on whether k is even or odd, minSet will always contain the first set of numbers (smaller) of window k and maxSet will contain the second set of numbers (larger).

As we move our window, we will remove elements from either of the sets (log n) and add a new element (log n) maintaining the minSet and maxSet rule specified above.

Below is the implementation of the above approach:

filter_none

edit
close

play_arrow

link
brightness_4
code

// Java implementation of the approach
import java.util.TreeSet;
  
public class GFG {
  
    // Pair class for the value and its index
    static class Pair implements Comparable<Pair> {
        private int value, index;
  
        // Constructor
        public Pair(int v, int p)
        {
            value = v;
            index = p;
        }
  
        // This method will be used by the treeset to
        // search a value by index and setting the tree
        // nodes (left or right)
        @Override
        public int compareTo(Pair o)
        {
  
            // Two nodes are equal only when
            // their indices are same
            if (index == o.index) {
                return 0;
            }
            else if (value == o.value) {
                return Integer.compare(index, o.index);
            }
            else {
                return Integer.compare(value, o.value);
            }
        }
  
        // Function to return the value
        // of the current object
        public int value()
        {
            return value;
        }
  
        // Update the value and the position
        // for the same object to save space
        public void renew(int v, int p)
        {
            value = v;
            index = p;
        }
  
        @Override
        public String toString()
        {
            return String.format("(%d, %d)", value, index);
        }
    }
  
    // Function to print the median for the current window
    static void printMedian(TreeSet<Pair> minSet,
                            TreeSet<Pair> maxSet, int window)
    {
  
        // If the window size is even then the
        // median will be the average of the
        // two middle elements
        if (window % 2 == 0) {
            System.out.print((minSet.last().value()
                              + maxSet.first().value())
                             / 2.0);
            System.out.print(" ");
        }
  
        // Else it will be the middle element
        else {
            System.out.print(minSet.size() > maxSet.size()
                                 ? minSet.last().value()
                                 : maxSet.first().value());
            System.out.print(" ");
        }
    }
  
    // Function to find the median
    // of every window of size k
    static void findMedian(int arr[], int k)
    {
        TreeSet<Pair> minSet = new TreeSet<>();
        TreeSet<Pair> maxSet = new TreeSet<>();
  
        // To hold the pairs, we will keep renewing
        // these instead of creating the new pairs
        Pair[] windowPairs = new Pair[k];
  
        for (int i = 0; i < k; i++) {
            windowPairs[i] = new Pair(arr[i], i);
        }
  
        // Add k/2 items to maxSet
        for (int i = 0; i < k / 2; i++) {
            maxSet.add(windowPairs[i]);
        }
  
        for (int i = k / 2; i < k; i++) {
  
            // Below logic is to maintain the
            // maxSet and the minSet criteria
            if (arr[i] < maxSet.first().value()) {
                minSet.add(windowPairs[i]);
            }
            else {
                minSet.add(maxSet.pollFirst());
                maxSet.add(windowPairs[i]);
            }
        }
  
        printMedian(minSet, maxSet, k);
  
        for (int i = k; i < arr.length; i++) {
  
            // Get the pair at the start of the window, this
            // will reset to 0 at every k, 2k, 3k, ...
            Pair temp = windowPairs[i % k];
            if (temp.value() <= minSet.last().value()) {
  
                // Remove the starting pair of the window
                minSet.remove(temp);
  
                // Renew window start to new window end
                temp.renew(arr[i], i);
  
                // Below logic is to maintain the
                // maxSet and the minSet criteria
                if (temp.value() < maxSet.first().value()) {
                    minSet.add(temp);
                }
                else {
                    minSet.add(maxSet.pollFirst());
                    maxSet.add(temp);
                }
            }
            else {
                maxSet.remove(temp);
                temp.renew(arr[i], i);
  
                // Below logic is to maintain the
                // maxSet and the minSet criteria
                if (temp.value() > minSet.last().value()) {
                    maxSet.add(temp);
                }
                else {
                    maxSet.add(minSet.pollLast());
                    minSet.add(temp);
                }
            }
  
            printMedian(minSet, maxSet, k);
        }
    }
  
    // Driver code
    public static void main(String[] args)
    {
        int[] arr = new int[] { 0, 9, 1, 8, 2,
                                7, 3, 6, 4, 5 };
        int k = 3;
  
        findMedian(arr, k);
    }
}

chevron_right


Output:

1 8 2 7 3 6 4 5



My Personal Notes arrow_drop_up

Check out this Author's contributed articles.

If you like GeeksforGeeks and would like to contribute, you can also write an article using contribute.geeksforgeeks.org or mail your article to contribute@geeksforgeeks.org. See your article appearing on the GeeksforGeeks main page and help other Geeks.

Please Improve this article if you find anything incorrect by clicking on the "Improve Article" button below.