I want to generate all distinct permutations of array of integers. The array may contain duplicates. but i want to generate all distinct permutations. I have tried next permutation and recursive methods which tend to be very slow. Please suggest.
Asked
Active
Viewed 2,462 times
1 Answers
1
There are n! different permutations of n elements. Generating a single permutation is cost n (strictly) so the minimum cost of any permutation generation algorithm would be O(n*n!)
Steinhaus–Johnson–Trotter algorithm is one of those algorithms. There are improvements like Shimon Even's and other algorithms like Heap's but none get them under O(n*n!)
Googling "permutation algorithm" gets several different algorithms you can implement, although most use recursion and that means another stack step. Steinhaus–Johnson–Trotter is defined as iterative, so shouldn't get that problem.
Here's a Java implementation
import java.util.Arrays;
import java.util.Iterator;
/**
* this implementation is based in Steinhaus–Johnson–Trotter algorithm and
* Shimon Even's improvement;
*
* @see https
* ://en.wikipedia.org/wiki/Steinhaus%E2%80%93Johnson%E2%80%93Trotter_algorithm
*
*/
public class Permutations implements Iterator<int[]> {
/**
* direction[i] = -1 if the element i has to move to the left, +1 to the
* right, 0 if it does not need to move
*/
private int[] direction;
/**
* inversePermutation[i] is the position of element i in permutation; It's
* called inverse permutation because if p2 is the inverse permutation of
* p1, then p1 is the inverse permutation of p2
*/
private int[] inversePermutation;
/**
* current permutation
*/
private int[] permutation;
/**
* @param numElements
* >= 1
*/
public Permutations(int numElements) {
// initial permutation
permutation = new int[numElements];
for (int i = 0; i < numElements; i++) {
permutation[i] = i;
}
// the support elements
inversePermutation = Arrays.copyOf(permutation, numElements);
direction = new int[numElements];
Arrays.fill(direction, -1);
direction[0] = 0;
}
/**
* Swaps the elements in array at positions i1 and i2
*
* @param array
* @param i1
* @param i2
*/
private static void swap(int[] array, int i1, int i2) {
int temp = array[i1];
array[i1] = array[i2];
array[i2] = temp;
}
/**
* prepares permutation to be the next one to return
*/
private void buildNextPermutation() {
// find the largest element with a nonzero direction, and swaps it in
// the indicated direction
int index = -1;
for (int i = 0; i < direction.length; i++) {
if (direction[permutation[i]] != 0
&& (index < 0 || permutation[index] < permutation[i])) {
index = i;
}
}
if (index < 0) {
// there are no more permutations
permutation = null;
} else {
// element we're moving
int chosenElement = permutation[index];
// direction we're moving
int dir = direction[chosenElement];
// index2 is the new position of chosenElement
int index2 = index + dir;
// we'll swap positions elements permutation[index] and
// permutation[index2] in permutation, to keep inversePermutation we
// have to swap inversePermutation's elements at index
// permutation[index] and permutation[index2]
swap(inversePermutation, permutation[index], permutation[index2]);
swap(permutation, index, index2);
// update directions
if (index2 == 0 || index2 == permutation.length - 1
|| permutation[index2 + dir] > permutation[index2]) {
// direction of chosen element
direction[chosenElement] = 0;
}
// all elements greater that chosenElement set its direction to +1
// if they're before index-1 or -1 if they're after
for (int i = chosenElement + 1; i < direction.length; i++) {
if (inversePermutation[i] > index2) {
direction[i] = -1;
} else {
direction[i] = 1;
}
}
}
}
@Override
public boolean hasNext() {
return permutation != null;
}
@Override
public int[] next() {
int[] result = Arrays.copyOf(permutation, permutation.length);
buildNextPermutation();
return result;
}
}

dtortola
- 768
- 4
- 6
-
If you use the Steinhaus-Johnson-Trotter permutation algorithm to re-order an existing array to step from one permutation to the next, by making a swap of two adjacent elements, (as opposed to generating a new array at each iteration) then it has a maximum cost of `O(N)` for a single step but the average cost per step is `O(1)` if you are stepping through all the `N!` permutations. This gives a total time for iteration over all the permutations that is `O(N!)`. See [this Thesis (page 94)](https://archive.org/details/PlanarityTestingByPathAddition) for a proof. – MT0 Jan 20 '16 at 10:39