Python-NumPy slice differently each line

Question

I have a small problem, as I am learning Python. I am trying to slice a 2D-array in a particular way : taking one item over two, but at each line we begin at a different index, for example if we have a = np.reshape(np.arange(16),(4,4)) , so

>>print(a) = [[ 0  1  2  3]
              [ 4  5  6  7]
              [ 8  9 10 11]
              [12 13 14 15]]

We would like to end with

>>print(new_a) = [[ 0  2]
                  [ 5  7]
                  [ 8 10]
                  [13 15]]

I am sure that it is not too complicated, but I couldn't find the answer :( (I now how to slice a np-array, just not how to change each row)

Thank you and have a nice day !

Paul Panzer · Accepted Answer · 2018-10-20 09:49:06Z

2

As long as the dimensions of a are even, one can get this specific (checkerboard) pattern with np.einsum:

>>> np.einsum('jiki->jik', a.reshape(2, 2, 2, 2)).reshape(4, 2)
array([[ 0,  2],
       [ 5,  7],
       [ 8, 10],
       [13, 15]])

or, more generally

>>> a = np.arange(40).reshape(4, 10)
>>> np.einsum('jiki->jik', a.reshape(a.shape[0]//2, 2, -1, 2)).reshape(a.shape[0], -1)
array([[ 0,  2,  4,  6,  8],
       [11, 13, 15, 17, 19],
       [20, 22, 24, 26, 28],
       [31, 33, 35, 37, 39]])

edited Oct 20, 2018 at 9:49

answered Oct 20, 2018 at 9:42

Paul Panzer

53.3k3 gold badges60 silver badges103 bronze badges

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

Comments

charlie6411 · Accepted Answer · 2018-10-20 09:33:39Z

0

You can use slicing with steps. Do this from 0 and from 1 with step size 2. This will give you two separate arrays, you can then do the work and join them back.

I am pretty new to numpy so there may be a better approach.

answered Oct 20, 2018 at 9:33

charlie6411

1067 bronze badges

4 Comments

Sanimys Over a year ago

Seems good, but how can you join them "one row over two" ? I mean, if we have b = a[::2,::2] and c[1::2,1::2], we would want a_new = b[0] then c[0], then b[1] and then c[1], and so on if a is big

charlie6411 Over a year ago

so do [0::2] ([ 0 1 2 3],[ 8 9 10 11]...] and [1::2] manipulate the data (can change rows and columns around if needed so with [0::2] you know you only want 0 and 2 with [1::2] you only want 1 and 3)..so get rid of stuff not using then can stack or concat and reshape if required...but I think @Paul Panzer answer is probably better

charlie6411 Over a year ago

..so if you understand slicing and want to keep to that you can do above to get your [ 0 1 2 3],[ 8 9 10 11] etc then flatten slice again every other one to get 0,2,8,10... and reshape

charlie6411 Over a year ago

do that with the second array [ 4 5 6 7],[12 13 14 15].. gives you two arrays which you have reshaped to be [0,2],[8,10].. and [5,7],[13,15] then append one to the other and sort if required

deadvoid · Accepted Answer · 2018-10-20 09:42:46Z

0

b = a.take([0,2,5,7,8,10,13,15]).reshape(4,2)
print(b)

array([[ 0,  2],
       [ 5,  7],
       [ 8, 10],
       [13, 15]])

answered Oct 20, 2018 at 9:42

deadvoid

1,3101 gold badge11 silver badges19 bronze badges

2 Comments

Sanimys Over a year ago

Works well in this case, but my matrix a was just an example, I work with a 5494x3666 image, so... :P

deadvoid Over a year ago

well, the numbers are also the exact indexes of your array, so what you can do is generate those indexes numbers in a separate function, then use the resulting list as np.array.take() argument.

Collectives™ on Stack Overflow

Python-NumPy slice differently each line

3 Answers 3

Comments

4 Comments

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

Comments

4 Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related