Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

transpose/rotate a block of a matrix in python

i have a 6x6 matrix as a list of lists in python. The matrix is divided into 4 square blocks of size 3x3. I want a way to take a transpose of only 1 block. I can do it using the traditional method of going through each element and copying it into another array and back and so on but I want to see if there is a better way, (transposing a matrix in python can be done in one line using the zip method)

for eg this is the representation of the matrix and its blocks

 block 1  block 2
+-------+-------+
| . . . | . . . |
| . . 2 | 1 . . |
| . . . | . . . |
+-------+-------+
| . . . | . . . |
| . . . | . . . |
| . 1 . | . . . |
+-------+-------+
 block 3  block 4

and rotate(3, right) should result in this

 block 1  block 2
+-------+-------+
| . . . | . . . |
| . . 2 | 1 . . |
| . . . | . . . |
+-------+-------+
| . . . | . . . |
| 1 . . | . . . |
| . . . | . . . |
+-------+-------+
 block 3  block 4

I want to find a method that takes in a block number and rotates only that block left or right. Is there any easy way to do it?

like image 855
randomThought Avatar asked Nov 10 '11 22:11

randomThought


2 Answers

Building on Sven Marnach's idea to use np.rot90, here is a version which rotates the quadrant clockwise (as requested?). In the key step

block3[:] = np.rot90(block3.copy(),-1)

a copy() is used on the right-hand side (RHS). Without the copy(), as values are assigned to block3, the underlying data used on the RHS is also changed. This muddles the values used in subsquent assignments. Without the copy(), multiple same values are spread about block3.

I don't see a way to do this operation without a copy.

import numpy as np
a = np.arange(36).reshape(6, 6)
print(a)
# [[ 0  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 26 27 28 29]
#  [30 31 32 33 34 35]]
block3 = a[3:6, 0:3]

# To rotate counterclockwise
block3[:] = np.rot90(block3.copy())
print(a)
# [[ 0  1  2  3  4  5]
#  [ 6  7  8  9 10 11]
#  [12 13 14 15 16 17]
#  [20 26 32 21 22 23]
#  [19 25 31 27 28 29]
#  [18 24 30 33 34 35]]

# To rotate clockwise
a = np.arange(36).reshape(6, 6)
block3 = a[3:6, 0:3]
block3[:] = np.rot90(block3.copy(),-1)
print(a)
# [[ 0  1  2  3  4  5]
#  [ 6  7  8  9 10 11]
#  [12 13 14 15 16 17]
#  [30 24 18 21 22 23]
#  [31 25 19 27 28 29]
#  [32 26 20 33 34 35]]
like image 167
unutbu Avatar answered Oct 05 '22 23:10

unutbu


For what it's worth, here's how simple this in in NumPy:

>>> a = numpy.arange(36).reshape(6, 6)
>>> a
array([[ 0,  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, 26, 27, 28, 29],
       [30, 31, 32, 33, 34, 35]])
>>> block3 = a[3:6, 0:3]
>>> block3[:] = numpy.rot90(block3, 1).copy()
>>> a
array([[ 0,  1,  2,  3,  4,  5],
       [ 6,  7,  8,  9, 10, 11],
       [12, 13, 14, 15, 16, 17],
       [20, 26, 32, 21, 22, 23],
       [26, 25, 31, 27, 28, 29],
       [20, 26, 20, 33, 34, 35]])
like image 33
Sven Marnach Avatar answered Oct 06 '22 00:10

Sven Marnach