Expanding a numpy array of matrices with zeros

Question

What's the most efficient way to pad an array of matrices with zeros?

example:

# Lets construct an array of 2 matrices from 3 arrays of vectors
import numpy as np

A = np.array([[0,1,2],[3,4,5]])       # 2 vectors
B = np.array([[6,7,8],[9,10,11]])     # 2 vectors
C = np.array([[12,13,14],[15,16,17]]) # 2 vectors
M = np.dstack((A,B,C))
'''
# Result: array([[[ 0,  6, 12],
                  [ 1,  7, 13],
                  [ 2,  8, 14]],

                 [[ 3,  9, 15],
                  [ 4, 10, 16],
                  [ 5, 11, 17]]]) #
'''

I want to add a column and/or a row of zeros to every matrix element in the array such as:

'''
# Result: array([[[ 0,  6, 12, 0],
              [ 1,  7, 13, 0],
              [ 2,  8, 14, 0],
              [ 0,  0,  0, 0]],

             [[ 3,  9, 15, 0],
              [ 4, 10, 16, 0],
              [ 5, 11, 17, 0]
              [ 0,  0,  0, 0]]]) #
'''

hpaulj · Accepted Answer · 2016-03-01 04:34:08Z

2

np.pad will work, but for this case it is overkill. We can do it directly with:

A sample 3d array (different dimensions make changes more obvious)

In [408]: M=np.arange(2*3*4).reshape((2,3,4))    
In [409]: M
Out[409]: 
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]]])

A blank array of the desired target shape

In [410]: M1=np.zeros((2,4,5),M.dtype)

Copy values from M to the target in the right slice range.

In [411]: M1[:,:-1,:-1]=M
In [412]: M1
Out[412]: 
array([[[ 0,  1,  2,  3,  0],
        [ 4,  5,  6,  7,  0],
        [ 8,  9, 10, 11,  0],
        [ 0,  0,  0,  0,  0]],

       [[12, 13, 14, 15,  0],
        [16, 17, 18, 19,  0],
        [20, 21, 22, 23,  0],
        [ 0,  0,  0,  0,  0]]])

A copy like this is required. There's no way of expanding the size of M itself. pad returns a new array as well, having performed a general version of this allocate and copy. So there isn't much of an efficiency issue.

You could also concatenate (or 'append') a 0 row or column in the right dimensions. But what I've illustrated does it in one step.

answered Mar 1, 2016 at 4:34

hpaulj

233k14 gold badges260 silver badges392 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Fnord Over a year ago

nice, this is about 3x faster than np.pad on my use case (over 1 million matrices)

Suever · Accepted Answer · 2016-03-01 03:50:27Z

1

You're going to want to use numpy.pad to do this. You can use the second input argument to specify the amount to pad before and after the data in each of the dimensions of the array. Then specify a constant padding type where the padding value is 0 (the default).

result = numpy.pad(M, ((0,0),(0,1),(0,1)), 'constant', constant_values=0)
array([[[ 0,  6, 12,  0],
        [ 1,  7, 13,  0],
        [ 2,  8, 14,  0],
        [ 0,  0,  0,  0]],

       [[ 3,  9, 15,  0],
        [ 4, 10, 16,  0],
        [ 5, 11, 17,  0],
        [ 0,  0,  0,  0]]])

answered Mar 1, 2016 at 3:50

Suever

65.6k14 gold badges91 silver badges104 bronze badges

Collectives™ on Stack Overflow

Expanding a numpy array of matrices with zeros

2 Answers 2

1 Comment

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related