Deleting rows and columns (of each matrix ) from a collection of matrices using a for-loop-CodePudding

I have a collection of (6 by 6) matrices, In my code, I may get 5 to 6 matrices (of order (6 by 6).)

I wanted to delete certain rows and columns in each of these 5 to 6 matrices I have.

so I ran a for-loop

suppose for now, in that collection I have only "one" matrix (a_matrix of order 6 by 6).

Each matrix that comes into this for-loop should get reduced to 2 by 2 matrices shown below. (desired output)

this is my failed attempt, here a_matrix will not work as there is only a single 6 by 6 matrix in my collection.

it is showing an error as

ValueError: could not broadcast input array from shape (5) into shape (6)

import numpy as np


a_matrix = np.array (      [  [   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,               36      ]] )
                        
               
total_no_of_matrices = 1                  


for i in range(total_no_of_matrices):
    a_matrix[i] = np.delete(a_matrix[i], 0, 0) 
    a_matrix[i] = np.delete(a_matrix[i], 0, 0)
    a_matrix[i] = np.delete(a_matrix[i], 1, 0)
    a_matrix[i] = np.delete(a_matrix[i], 1, 0)


print(a_matrix)

desired output- (after deletion of certain rows and columns)

a_matrix = [[15, 18],
            [33, 36]]

CodePudding user response：

The issue is that numpy makes arrays as static as possible. So, if you declare an array to be nx6x6 you can't suddenly tell it that you want matrix 3 to be 2x2. The better way is to declare a new array that will hold the smaller matrices. There are some cases where you can make a numpy array dynamic, but it is usually very inefficient and usually you preallocate the arrays anyway -- if you can't, a numpy array is probably not the right object to use. Here's how I might approach this:

import numpy as np


a_matrix = np.arange(1, 6 * 6   1).reshape((6, 6))  
                        
               
if a_matrix.ndim == 2:
    a_matrix = a_matrix.reshape((1, a_matrix.shape[0], a_matrix.shape[1]))
total_no_of_matrices = a_matrix.shape[0]
reduced_matrices = np.zeros((total_no_of_matrices, 2, 2))


rows_to_delete = [0, 1, 3, 4]
cols_to_delete = [0, 1, 3, 4]
for i in range(total_no_of_matrices):
    temp_matrix = np.delete(a_matrix[i, :, :], rows_to_delete, axis=0) 
    temp_matrix = np.delete(temp_matrix, cols_to_delete, axis=1)
    reduced_matrices[i, :, :] = temp_matrix


print(a_matrix)
print(reduced_matrices)

Also, if you want to get really fancy, you can use advanced indexing. But sometimes I find it hard to follow, and I use numpy all the time. So if anyone reads the code maybe stay away from it, but here is another option

another_way = a_matrix[:, np.array([[2, 2], [5, 5]]), np.array([[2, 5], [2, 5]])]
print(another_way)

CodePudding user response：

Say you have 20 such matrices, so (20,6,6) is their shape. You can use indexing like matrices[:, 2:6:3, 2:6:3] to get all 2 x 2 reduced matrices.

import numpy as np

matrices = np.arange(1, 20*6*6 1).reshape((20, 6, 6))
reduced_matrices = matrices[:, 2:6:3, 2:6:3]

See the first one for example:

reduced_matrices[0]
[[15 18]
 [33 36]]

If you like using loops, then a simple comprehension could do the same as well:

reduced_matrices = [ m[2:6:3,2:6:3] for m in matrices ]