Hi All,
I found the comparison between these two random shuffling algorithms to be interesting.
Consider two shuffling algorithms
SHUFFLE 1
shuffle(A[1 … n]) {
for i = 1 to n {
// Find a random integer between 1 and n inclusive
int rand= RANDOM(1,n);
swap A[i] with A[rand];
}
}
SHUFFLE 2
shuffle(A[1 … n]) {
for i = 1 to n {
// Find a random integer between i and n inclusive
int rand= RANDOM(i,n);
swap A[i] with A[rand];
}
}
How do these two shuffling algorithms compare against each other?
Which of these two is a perfect shuffling algorithm?
Consider an array with distinct elements A[1 … n]
A perfectly unbiased shuffle algorithm would randomly shuffle all elements in the array such that after shuffling:
1.The probability that the shuffling operation would produce any particular permutation of the original array is the same for all permutations (i.e.) since there are n! permutations, the probability that the shuffle operation would produce any particular permutation is 1/n!
2.For any element e in the array and for any position j (1<= j <= n), the probability that the element would end up in position A[j] is 1/n
Simulating Shuffle 1 and Shuffle 2 clearly proves that Shuffle 1 is biased while Shuffle 2 is unbiased
Can we prove that Shuffle 2 will produce an unbiased shuffle in all cases?
For any element e, the probability that it will be shuffled into the first position
= probability that it is selected for swapping when i = 1
= 1/n
For any element e, the probability that it will be shuffled into the second position
= probability that it is NOT selected for the first position x probability that it is selected for swapping when i = 2
= (n-1)/n x 1/(n-1)
= 1/n
…
For any element e, the probability that it will be shuffled into any particular position = 1/n