O(1) AVERAGE Time solution -- when assuming RANDOM input


  • -2
    J

    Many people have used binary search to guarantee O(nlgn) time--which I never thought about as I stopped when I got the following O(n) solution.
    The general idea is to do one pass from head to tail, check the condition num[i-1] < num[i] > num[i+1], return the first i that satisfies. So at first glance, it's a O(n) time, as the only peak may be close to the tail.

    public class Solution {
        public int findPeakElement(int[] num) {
            if(num == null) return -1;
            if(num.length == 1) return 0;
            
            if(num[0] > num[1]) return 0;
            if(num[num.length - 1] > num[num.length - 2]) return num.length - 1;
            
            for(int i = num.length - 2; i > 0; i --){
                if(num[i] > num[i+1] && num[i] > num[i-1]) return i;
            }
            return -1;
        }
    }
    

    However, here comes something interesting. After a second thought, I noticed when assuming random input, my solution is actually constant time in average. Here's why:

    Assumption: P(num[i] < num[i+1]) = P(num[i] > num[i+1]) = 1/2 -- assuming random input
    Notation: use p(x) to represent the chance of program finding a peak after checking x indices.
    p(1) = P(num[0] > num[1]) = 1/2;
    p(2) = (1 - p(1)) * P(num[1] > num[2]) = 1/2 * 1/2 = 1/4;
    p(3) = 1/8
    ...
    p(x) = 1/2^x
    so on average, the # of indices to check before finding a peak is 1p(1) + 2p(2) + 3*p(3) + .... = 2

    so actually, this O(n) bound of first glance is pretty "loose" in some sense, as the average time is O(2)!

    However, this analysis kind of relies on the "randomness" of the input, but I think it's still something worth mentioning.

    Having fun solving!


  • 1
    F

    You may not see this sentence : "Your solution should be in logarithmic complexity.", which means we should make the time complexity be O(logn) instead of O(n).
    And for your solution, I think the average time complexity is (1+2+3+.....+n-1) / n = (n-1)/2, but not O(2).


  • 0
    L

    1,2,3,4,..,n-1 don't have the same probability to be the answer, when the data is random, the above solution is a amotized O(1) one.


  • 0
    U

    if the input data is "pure random", there is no need to calculate, just pick two return the bigger one.


Log in to reply
 

Looks like your connection to LeetCode Discuss was lost, please wait while we try to reconnect.