k-th Largest Element (Quickselect)

medium Searching and Sorting

Problem

Print the k-th biggest number of the array (counting duplicates: in 5 5 3 the 2nd biggest is 5). Quickselect finds it without sorting the whole array: it partitions around a pivot and continues only in the side that holds the answer.

Constraints

  • 1 <= n <= 10^5
  • 1 <= k <= n

Examples

Input

6
3 2 1 5 6 4
2

Output

5

Solutions

Best approachTime O(n) averageSpace O(n)

After one partition the pivot stands at its final sorted place. If that is the target index we are done; otherwise only one side needs more work. The average cost is O(n); a bad pivot can make it O(n^2).

Same code works in Java810172125Basic features only, so nothing changes between versions.
/*
 * Problem : k-th largest.
 * Input   : n, the array, then k.
 * Output  : The number.
 * Approach: Quickselect.
 * Time    : O(n) average
 * Space   : O(n)
 * Java    : 8 or newer.
 */
import java.util.Scanner;

public class Main {
    // This method solves the task. It receives the input and returns the answer.
    static int solve(int[] a, int k) {
        // Step 1: Work on a copy. In ascending order the answer sits at index target = n - k.
        int[] data = a.clone();
        int target = data.length - k;
        int low = 0;
        int high = data.length - 1;
        while (low < high) {
            // Step 2: Partition: move the middle value to the end as the pivot. Smaller numbers go to the left side.
            int middle = low + (high - low) / 2;
            int temp = data[middle];
            data[middle] = data[high];
            data[high] = temp;
            int pivot = data[high];
            int store = low;
            for (int i = low; i < high; i++) {
                if (data[i] < pivot) {
                    temp = data[i];
                    data[i] = data[store];
                    data[store] = temp;
                    store++;
                }
            }
            temp = data[store];
            data[store] = data[high];
            data[high] = temp;
            // Step 3: The pivot is now at its final place "store". Continue only on the side that holds the target.
            if (store == target) {
                return data[store];
            } else if (store < target) {
                low = store + 1;
            } else {
                high = store - 1;
            }
        }
        return data[target];
    }

    public static void main(String[] args) {
        Scanner scanner = new Scanner(System.in);
        int size = scanner.nextInt();
        int[] a = new int[size];
        for (int i = 0; i < size; i++) {
            a[i] = scanner.nextInt();
        }
        int k = scanner.nextInt();
        System.out.println(solve(a, k));
    }
}
Hints (2)
  1. Where does the pivot end up after partition?
  2. Why do we only follow one side?