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

user_18728568 0
Viewed 57

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]]
indexing matrix arrays python loops
2 Answers

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)
user_80562480

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 ]
user_39431700
Related