Inspired by Raymond Chen's post, say you have a 4x4 two dimensional array, write a function that rotates it 90 degrees. Raymond links to a solution in pseudo code, but I'd like to see some real world stuff.
[1][2][3][4]
[5][6][7][8]
[9][0][1][2]
[3][4][5][6]
Becomes:
[3][9][5][1]
[4][0][6][2]
[5][1][7][3]
[6][2][8][4]
Update: Nick's answer is the most straightforward, but is there a way to do it better than n^2? What if the matrix was 10000x10000?
Here it is in C#
int[,] array = new int[4,4] {
{ 1,2,3,4 },
{ 5,6,7,8 },
{ 9,0,1,2 },
{ 3,4,5,6 }
};
int[,] rotated = RotateMatrix(array, 4);
static int[,] RotateMatrix(int[,] matrix, int n) {
int[,] ret = new int[n, n];
for (int i = 0; i < n; ++i) {
for (int j = 0; j < n; ++j) {
ret[i, j] = matrix[n - j - 1, i];
}
}
return ret;
}
Python:
rotated = list(zip(*original[::-1]))
and counterclockwise:
rotated_ccw = list(zip(*original))[::-1]
How this works:
zip(*original) will swap axes of 2d arrays by stacking corresponding items from lists into new lists. (The * operator tells the function to distribute the contained lists into arguments)
>>> list(zip(*[[1,2,3],[4,5,6],[7,8,9]]))
[[1,4,7],[2,5,8],[3,6,9]]
The [::-1] statement reverses array elements (please see Extended Slices or this question):
>>> [[1,2,3],[4,5,6],[7,8,9]][::-1]
[[7,8,9],[4,5,6],[1,2,3]]
Finally, combining the two will result in the rotation transformation.
The change in placement of [::-1] will reverse lists in different levels of the matrix.
Here is one that does the rotation in place instead of using a completely new array to hold the result. I've left off initialization of the array and printing it out. This only works for square arrays but they can be of any size. Memory overhead is equal to the size of one element of the array so you can do the rotation of as large an array as you want.
int a[4][4];
int n = 4;
int tmp;
for (int i = 0; i < n / 2; i++)
{
for (int j = i; j < n - i - 1; j++)
{
tmp = a[i][j];
a[i][j] = a[j][n-i-1];
a[j][n-i-1] = a[n-i-1][n-j-1];
a[n-i-1][n-j-1] = a[n-j-1][i];
a[n-j-1][i] = tmp;
}
}
There are tons of good code here but I just want to show what's going on geometrically so you can understand the code logic a little better. Here is how I would approach this.
first of all, do not confuse this with transposition which is very easy..
the basica idea is to treat it as layers and we rotate one layer at a time..
say we have a 4x4
1 2 3 4
5 6 7 8
9 10 11 12
13 14 15 16
after we rotate it clockwise by 90 we get
13 9 5 1
14 10 6 2
15 11 7 3
16 12 8 4
so let's decompose this, first we rotate the 4 corners essentially
1 4
13 16
then we rotate the following diamond which is sort of askew
2
8
9
15
and then the 2nd skewed diamond
3
5
12
14
so that takes care of the outer edge so essentially we do that one shell at a time until
finally the middle square (or if it's odd just the final element which does not move)
6 7
10 11
so now let's figure out the indices of each layer, assume we always work with the outermost layer, we are doing
[0,0] -> [0,n-1], [0,n-1] -> [n-1,n-1], [n-1,n-1] -> [n-1,0], and [n-1,0] -> [0,0]
[0,1] -> [1,n-1], [1,n-2] -> [n-1,n-2], [n-1,n-2] -> [n-2,0], and [n-2,0] -> [0,1]
[0,2] -> [2,n-2], [2,n-2] -> [n-1,n-3], [n-1,n-3] -> [n-3,0], and [n-3,0] -> [0,2]
so on and so on until we are halfway through the edge
so in general the pattern is
[0,i] -> [i,n-i], [i,n-i] -> [n-1,n-(i+1)], [n-1,n-(i+1)] -> [n-(i+1),0], and [n-(i+1),0] to [0,i]
As I said in my previous post, here's some code in C# that implements an O(1) matrix rotation for any size matrix. For brevity and readability there's no error checking or range checking. The code:
static void Main (string [] args)
{
int [,]
// create an arbitrary matrix
m = {{0, 1}, {2, 3}, {4, 5}};
Matrix
// create wrappers for the data
m1 = new Matrix (m),
m2 = new Matrix (m),
m3 = new Matrix (m);
// rotate the matricies in various ways - all are O(1)
m1.RotateClockwise90 ();
m2.Rotate180 ();
m3.RotateAnitclockwise90 ();
// output the result of transforms
System.Diagnostics.Trace.WriteLine (m1.ToString ());
System.Diagnostics.Trace.WriteLine (m2.ToString ());
System.Diagnostics.Trace.WriteLine (m3.ToString ());
}
class Matrix
{
enum Rotation
{
None,
Clockwise90,
Clockwise180,
Clockwise270
}
public Matrix (int [,] matrix)
{
m_matrix = matrix;
m_rotation = Rotation.None;
}
// the transformation routines
public void RotateClockwise90 ()
{
m_rotation = (Rotation) (((int) m_rotation + 1) & 3);
}
public void Rotate180 ()
{
m_rotation = (Rotation) (((int) m_rotation + 2) & 3);
}
public void RotateAnitclockwise90 ()
{
m_rotation = (Rotation) (((int) m_rotation + 3) & 3);
}
// accessor property to make class look like a two dimensional array
public int this [int row, int column]
{
get
{
int
value = 0;
switch (m_rotation)
{
case Rotation.None:
value = m_matrix [row, column];
break;
case Rotation.Clockwise90:
value = m_matrix [m_matrix.GetUpperBound (0) - column, row];
break;
case Rotation.Clockwise180:
value = m_matrix [m_matrix.GetUpperBound (0) - row, m_matrix.GetUpperBound (1) - column];
break;
case Rotation.Clockwise270:
value = m_matrix [column, m_matrix.GetUpperBound (1) - row];
break;
}
return value;
}
set
{
switch (m_rotation)
{
case Rotation.None:
m_matrix [row, column] = value;
break;
case Rotation.Clockwise90:
m_matrix [m_matrix.GetUpperBound (0) - column, row] = value;
break;
case Rotation.Clockwise180:
m_matrix [m_matrix.GetUpperBound (0) - row, m_matrix.GetUpperBound (1) - column] = value;
break;
case Rotation.Clockwise270:
m_matrix [column, m_matrix.GetUpperBound (1) - row] = value;
break;
}
}
}
// creates a string with the matrix values
public override string ToString ()
{
int
num_rows = 0,
num_columns = 0;
switch (m_rotation)
{
case Rotation.None:
case Rotation.Clockwise180:
num_rows = m_matrix.GetUpperBound (0);
num_columns = m_matrix.GetUpperBound (1);
break;
case Rotation.Clockwise90:
case Rotation.Clockwise270:
num_rows = m_matrix.GetUpperBound (1);
num_columns = m_matrix.GetUpperBound (0);
break;
}
StringBuilder
output = new StringBuilder ();
output.Append ("{");
for (int row = 0 ; row <= num_rows ; ++row)
{
if (row != 0)
{
output.Append (", ");
}
output.Append ("{");
for (int column = 0 ; column <= num_columns ; ++column)
{
if (column != 0)
{
output.Append (", ");
}
output.Append (this [row, column].ToString ());
}
output.Append ("}");
}
output.Append ("}");
return output.ToString ();
}
int [,]
// the original matrix
m_matrix;
Rotation
// the current view of the matrix
m_rotation;
}
OK, I'll put my hand up, it doesn't actually do any modifications to the original array when rotating. But, in an OO system that doesn't matter as long as the object looks like it's been rotated to the clients of the class. At the moment, the Matrix class uses references to the original array data so changing any value of m1 will also change m2 and m3. A small change to the constructor to create a new array and copy the values to it will sort that out.
Whilst rotating the data in place might be necessary (perhaps to update the physically stored representation), it becomes simpler and possibly more performant to add a layer of indirection onto the array access, perhaps an interface:
interface IReadableMatrix
{
int GetValue(int x, int y);
}
If your Matrix already implements this interface, then it can be rotated via a decorator class like this:
class RotatedMatrix : IReadableMatrix
{
private readonly IReadableMatrix _baseMatrix;
public RotatedMatrix(IReadableMatrix baseMatrix)
{
_baseMatrix = baseMatrix;
}
int GetValue(int x, int y)
{
// transpose x and y dimensions
return _baseMatrix(y, x);
}
}
Rotating +90/-90/180 degrees, flipping horizontally/vertically and scaling can all be achieved in this fashion as well.
Performance would need to be measured in your specific scenario. However the O(n^2) operation has now been replaced with an O(1) call. It's a virtual method call which is slower than direct array access, so it depends upon how frequently the rotated array is used after rotation. If it's used once, then this approach would definitely win. If it's rotated then used in a long-running system for days, then in-place rotation might perform better. It also depends whether you can accept the up-front cost.
As with all performance issues, measure, measure, measure!
This a better version of it in Java: I've made it for a matrix with a different width and height
public int[][] rotateMatrixRight(int[][] matrix)
{
/* W and H are already swapped */
int w = matrix.length;
int h = matrix[0].length;
int[][] ret = new int[h][w];
for (int i = 0; i < h; ++i) {
for (int j = 0; j < w; ++j) {
ret[i][j] = matrix[w - j - 1][i];
}
}
return ret;
}
public int[][] rotateMatrixLeft(int[][] matrix)
{
/* W and H are already swapped */
int w = matrix.length;
int h = matrix[0].length;
int[][] ret = new int[h][w];
for (int i = 0; i < h; ++i) {
for (int j = 0; j < w; ++j) {
ret[i][j] = matrix[j][h - i - 1];
}
}
return ret;
}
This code is based on Nick Berardi's post.
A couple of people have already put up examples which involve making a new array.
A few other things to consider:
(a) Instead of actually moving the data, simply traverse the "rotated" array differently.
(b) Doing the rotation in-place can be a little trickier. You'll need a bit of scratch place (probably roughly equal to one row or column in size). There's an ancient ACM paper about doing in-place transposes (http://doi.acm.org/10.1145/355719.355729), but their example code is nasty goto-laden FORTRAN.
Addendum:
http://doi.acm.org/10.1145/355611.355612 is another, supposedly superior, in-place transpose algorithm.
Nick's answer would work for an NxM array too with only a small modification (as opposed to an NxN).
string[,] orig = new string[n, m];
string[,] rot = new string[m, n];
...
for ( int i=0; i < n; i++ )
for ( int j=0; j < m; j++ )
rot[j, n - i - 1] = orig[i, j];
One way to think about this is that you have moved the center of the axis (0,0) from the top left corner to the top right corner. You're simply transposing from one to the other.
A common method to rotate a 2D array clockwise or anticlockwise.
1 2 3 7 8 9 7 4 1
4 5 6 => 4 5 6 => 8 5 2
7 8 9 1 2 3 9 6 3
void rotate(vector<vector<int> > &matrix) {
reverse(matrix.begin(), matrix.end());
for (int i = 0; i < matrix.size(); ++i) {
for (int j = i + 1; j < matrix[i].size(); ++j)
swap(matrix[i][j], matrix[j][i]);
}
}
1 2 3 3 2 1 3 6 9
4 5 6 => 6 5 4 => 2 5 8
7 8 9 9 8 7 1 4 7
void anti_rotate(vector<vector<int> > &matrix) {
for (auto vi : matrix) reverse(vi.begin(), vi.end());
for (int i = 0; i < matrix.size(); ++i) {
for (int j = i + 1; j < matrix[i].size(); ++j)
swap(matrix[i][j], matrix[j][i]);
}
}
Here's my Ruby version (note the values aren't displayed the same, but it still rotates as described).
def rotate(matrix)
result = []
4.times { |x|
result[x] = []
4.times { |y|
result[x][y] = matrix[y][3 - x]
}
}
result
end
matrix = []
matrix[0] = [1,2,3,4]
matrix[1] = [5,6,7,8]
matrix[2] = [9,0,1,2]
matrix[3] = [3,4,5,6]
def print_matrix(matrix)
4.times { |y|
4.times { |x|
print "#{matrix[x][y]} "
}
puts ""
}
end
print_matrix(matrix)
puts ""
print_matrix(rotate(matrix))
The output:
1 5 9 3
2 6 0 4
3 7 1 5
4 8 2 6
4 3 2 1
8 7 6 5
2 1 0 9
6 5 4 3
here's a in-space rotate method, by java, only for square. for non-square 2d array, you will have to create new array anyway.
private void rotateInSpace(int[][] arr) {
int z = arr.length;
for (int i = 0; i < z / 2; i++) {
for (int j = 0; j < (z / 2 + z % 2); j++) {
int x = i, y = j;
int temp = arr[x][y];
for (int k = 0; k < 4; k++) {
int temptemp = arr[y][z - x - 1];
arr[y][z - x - 1] = temp;
temp = temptemp;
int tempX = y;
y = z - x - 1;
x = tempX;
}
}
}
}
code to rotate any size 2d array by creating new array:
private int[][] rotate(int[][] arr) {
int width = arr[0].length;
int depth = arr.length;
int[][] re = new int[width][depth];
for (int i = 0; i < depth; i++) {
for (int j = 0; j < width; j++) {
re[j][depth - i - 1] = arr[i][j];
}
}
return re;
}
This is my implementation, in C, O(1) memory complexity, in place rotation, 90 degrees clockwise:
#include <stdio.h>
#define M_SIZE 5
static void initMatrix();
static void printMatrix();
static void rotateMatrix();
static int m[M_SIZE][M_SIZE];
int main(void){
initMatrix();
printMatrix();
rotateMatrix();
printMatrix();
return 0;
}
static void initMatrix(){
int i, j;
for(i = 0; i < M_SIZE; i++){
for(j = 0; j < M_SIZE; j++){
m[i][j] = M_SIZE*i + j + 1;
}
}
}
static void printMatrix(){
int i, j;
printf("Matrix\n");
for(i = 0; i < M_SIZE; i++){
for(j = 0; j < M_SIZE; j++){
printf("%02d ", m[i][j]);
}
printf("\n");
}
printf("\n");
}
static void rotateMatrix(){
int r, c;
for(r = 0; r < M_SIZE/2; r++){
for(c = r; c < M_SIZE - r - 1; c++){
int tmp = m[r][c];
m[r][c] = m[M_SIZE - c - 1][r];
m[M_SIZE - c - 1][r] = m[M_SIZE - r - 1][M_SIZE - c - 1];
m[M_SIZE - r - 1][M_SIZE - c - 1] = m[c][M_SIZE - r - 1];
m[c][M_SIZE - r - 1] = tmp;
}
}
}
In python:
import numpy as np
a = np.array(
[
[1, 2, 3, 4],
[5, 6, 7, 8],
[9, 0, 1, 2],
[3, 4, 5, 6]
]
)
print(a)
print(b[::-1, :].T)
@dagorym: Aw, man. I had been hanging onto this as a good "I'm bored, what can I ponder" puzzle. I came up with my in-place transposition code, but got here to find yours pretty much identical to mine...ah, well. Here it is in Ruby.
require 'pp'
n = 10
a = []
n.times { a << (1..n).to_a }
pp a
0.upto(n/2-1) do |i|
i.upto(n-i-2) do |j|
tmp = a[i][j]
a[i][j] = a[n-j-1][i]
a[n-j-1][i] = a[n-i-1][n-j-1]
a[n-i-1][n-j-1] = a[j][n-i-1]
a[j][n-i-1] = tmp
end
end
pp a
short normal[4][4] = {{8,4,7,5},{3,4,5,7},{9,5,5,6},{3,3,3,3}};
short rotated[4][4];
for (int r = 0; r < 4; ++r)
{
for (int c = 0; c < 4; ++c)
{
rotated[r][c] = normal[c][3-r];
}
}
Simple C++ method, tho there would be a big memory overhead in a big array.
#include <iostream>
#include <iomanip>
using namespace std;
const int SIZE=3;
void print(int a[][SIZE],int);
void rotate(int a[][SIZE],int);
void main()
{
int a[SIZE][SIZE]={{11,22,33},{44,55,66},{77,88,99}};
cout<<"the array befor rotate\n";
print(a,SIZE);
rotate( a,SIZE);
cout<<"the array after rotate\n";
print(a,SIZE);
cout<<endl;
}
void print(int a[][SIZE],int SIZE)
{
int i,j;
for(i=0;i<SIZE;i++)
for(j=0;j<SIZE;j++)
cout<<a[i][j]<<setw(4);
}
void rotate(int a[][SIZE],int SIZE)
{
int temp[3][3],i,j;
for(i=0;i<SIZE;i++)
for(j=0;j<SIZE/2.5;j++)
{
temp[i][j]= a[i][j];
a[i][j]= a[j][SIZE-i-1] ;
a[j][SIZE-i-1] =temp[i][j];
}
}
Based on the community wiki algorithm and this SO answer for transposing arrays, here is a Swift 4 version to rotate some 2D array 90 degrees counter-clockwise. This assumes matrix is a 2D array:
func rotate(matrix: [[Int]]) -> [[Int]] {
let transposedPoints = transpose(input: matrix)
let rotatedPoints = transposedPoints.map{ Array($0.reversed()) }
return rotatedPoints
}
fileprivate func transpose<T>(input: [[T]]) -> [[T]] {
if input.isEmpty { return [[T]]() }
let count = input[0].count
var out = [[T]](repeating: [T](), count: count)
for outer in input {
for (index, inner) in outer.enumerated() {
out[index].append(inner)
}
}
return out
}
This solution doesn't care square or rectangle dimension, you can rotate 4x5 or 5x4 or even 4x4, it doesn't care the size as well. Note that this implementation creates a new array every time you call rotate90 method, it doesn't mutate the original array at all.
public static void main(String[] args) {
int[][] a = new int[][] {
{ 1, 2, 3, 4 },
{ 5, 6, 7, 8 },
{ 9, 0, 1, 2 },
{ 3, 4, 5, 6 },
{ 7, 8, 9, 0 }
};
int[][] rotate180 = rotate90(rotate90(a));
print(rotate180);
}
static int[][] rotate90(int[][] a) {
int[][] ret = new int[a[0].length][a.length];
for (int i = 0; i < a.length; i++) {
for (int j = 0; j < a[i].length; j++) {
ret[j][a.length - i - 1] = a[i][j];
}
}
return ret;
}
static void print(int[][] array) {
for (int i = 0; i < array.length; i++) {
System.out.print("[");
for (int j = 0; j < array[i].length; j++) {
System.out.print(array[i][j]);
System.out.print(" ");
}
System.out.println("]");
}
}
I was able to do this with a single loop. The time complexity seems like O(K) where K is all items of the array. Here's how I did it in JavaScript:
First off, we represent the n^2 matrix with a single array. Then, iterate through it like this:
/**
* Rotates matrix 90 degrees clockwise
* @param arr: the source array
* @param n: the array side (array is square n^2)
*/
function rotate (arr, n) {
var rotated = [], indexes = []
for (var i = 0; i < arr.length; i++) {
if (i < n)
indexes[i] = i * n + (n - 1)
else
indexes[i] = indexes[i - n] - 1
rotated[indexes[i]] = arr[i]
}
return rotated
}
Basically, we transform the source array indexes:
[0,1,2,3,4,5,6,7,8] => [2,5,8,1,4,7,0,3,6]
Then, using this transformed indexes array, we place the actual values in the final rotated array.
Here are some test cases:
//n=3
rotate([
1, 2, 3,
4, 5, 6,
7, 8, 9], 3))
//result:
[7, 4, 1,
8, 5, 2,
9, 6, 3]
//n=4
rotate([
1, 2, 3, 4,
5, 6, 7, 8,
9, 10, 11, 12,
13, 14, 15, 16], 4))
//result:
[13, 9, 5, 1,
14, 10, 6, 2,
15, 11, 7, 3,
16, 12, 8, 4]
//n=5
rotate([
1, 2, 3, 4, 5,
6, 7, 8, 9, 10,
11, 12, 13, 14, 15,
16, 17, 18, 19, 20,
21, 22, 23, 24, 25], 5))
//result:
[21, 16, 11, 6, 1,
22, 17, 12, 7, 2,
23, 18, 13, 8, 3,
24, 19, 14, 9, 4,
25, 20, 15, 10, 5]
In Eigen (C++):
Eigen::Matrix2d mat;
mat << 1, 2,
3, 4;
std::cout << mat << "\n\n";
Eigen::Matrix2d r_plus_90 = mat.transpose().rowwise().reverse();
std::cout << r_plus_90 << "\n\n";
Eigen::Matrix2d r_minus_90 = mat.transpose().colwise().reverse();
std::cout << r_minus_90 << "\n\n";
Eigen::Matrix2d r_180 = mat.colwise().reverse().rowwise().reverse(); // +180 same as -180
std::cout << r_180 << "\n\n";
Output:
1 2
3 4
3 1
4 2
2 4
1 3
4 3
2 1
It is not possible to do it quicker than O(n^2) for in place rotation, for the reason that if we want to rotate the matrix, we have to touch all the n^2 element at least once, no matter what algorithm you are implementing.
Can be done recursively quite cleanly, here is my implementation in golang!
rotate nxn matrix in go golang recursively in place with no additional memory
func rot90(a [][]int) {
n := len(a)
if n == 1 {
return
}
for i := 0; i < n; i++ {
a[0][i], a[n-1-i][n-1] = a[n-1-i][n-1], a[0][i]
}
rot90(a[1:])
}