How to determine neighboring elements from a 1D array?

Question

I am working on a Finite element analysis code and I currently have 1D array listing the element density values like so:
x = np.ones(12) where the index is the element number 0, 1, 2, ..., 10, 11
The elements when plotted are like so:

0 - 3 - 6 - 9
1 - 4 - 7 - 10
2 - 5 - 8 - 11

I set the number of elements in the x and y direction (for this case 4 in the x and 3 in the y) however am having difficulty determining the surround elements. I need to find a way to determine the 3, 5 or 8 elements which surround a given elements. For example, if I select element 0 the surrounding elements are 1, 3, 4 or if I select element 6 the surrounding elements are 3, 4, 7, 9, 10 or if if I select element 7 the surround elements are 3, 4, 5, 6, 8, 9, 10, 11...
The end goal here would be to put in a radius and based on it determine the element numbers surrounding a selected element. Any advice or help with this would be greatly appreciated. For some reason I am unable to determine the logic to do this in python.

Wouldn't it make more sense to `.reshape((4,3))` and work with it in 2D? — Tim Roberts, Mar 30 '21 at 19:36
You can see the pattern, right? Up is `N-1`, down is `N+1`, left is `N-height`, right is `N+height`. — Tim Roberts, Mar 30 '21 at 19:38

wwii · Accepted Answer · 2021-03-31T17:26:48.230

determine the logic to do this

A 1-d array with 6 items has the indices - [0,1,2,3,4,5]
A proposed shape is 2 rows and 3 columns - M,N = 2,3.
for any item's index (i) its row and column are c,r = divmod(i,M)

The neighbor column,row indices will be

cplus,cminus = c + 1, c - 1
rplus, rminus = r + 1, r - 1
cplus,r
cminus,r
c,rplus
c,rminus
cplus,rplus
cplus,rminus
cminus,rplus
cminus,rminus

those 2d indices need to be converted to 1d indices with (col * M) + row

For example

[0,1,2,3,4,5]
M,N = 2,3
'''
0  2  4
1  3  5
'''

item 4's 2d index is c,r = divmod(4,M) --> (2,0) (col,row)
one of its neighbor's 2d index is c,rplus --> (2,1)
that neighbor's 1d index is (2 * M) + 1 --> 5
after converting the neighbors' 2d indices to 1d you will need to check for and discard some that don't make sense.
- item 4 is in the top-right a corner and doesn't have some neighbors like c,rminus which would be (2,-1) which does not make sense. Or cplus,r ... (3,0) which also does not make sense.

Caveat - I did NOT try to thoroughly test this.

Here is a function that returns a callable.

import operator
def get_neighbors(index, shape=(M,N)):
    '''Returns a callable.

    (M,N) --> (number_of_rows, number_of_columns)
    '''
    M, N = shape
    # 2d index
    c, r = divmod(index, M)
    print(f'2d index: {(c,r)}')
    # neighbors
    cplus, cminus = c + 1, c - 1
    rplus, rminus = r + 1, r - 1
    # dot product of (c,cplus,cminus) and (r,rplus,rminus)?
    neighbors = [
        (cminus, rminus),
        (cminus, r),
        (cminus, rplus),
        (c, rplus),
        (cplus, rplus),
        (cplus, r),
        (cplus, rminus),
        (c, rminus),
    ]
    # print(neighbors)

    # validate/filter
    neighbors = [
        (col, row) for col, row in neighbors if (0 <= col < N) and (0 <= row < M)
    ]
    # print(neighbors)

    # 1d indices
    one_d = [(col * M) + row for col,row in neighbors]
    # print(one_d)
    return operator.itemgetter(*one_d)

Try it out.

>>> a = ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't']

>>> M,N = 4,5    # nrows, ncols

'''
[['a' 'e' 'i' 'm' 'q']
 ['b' 'f' 'j' 'n' 'r']
 ['c' 'g' 'k' 'o' 's']
 ['d' 'h' 'l' 'p' 't']]
'''

>>> # i's neighbors
>>> q = get_neighbors(a.index('i'),(M,N))
2d index: (2, 0)
>>> q(a)
('e', 'f', 'j', 'n', 'm')
>>>
>>> # k's neighbors
>>> q = get_neighbors(a.index('k'),(M,N))
2d index: (2, 2)
>>> q(a)
('f', 'g', 'h', 'l', 'p', 'o', 'n', 'j')
>>>
>>> # q's neighbors
>>> q = get_neighbors(a.index('q'),(M,N))
2d index: (4, 0)
>>> q(a)
('m', 'n', 'r')
>>>

i's neighbors for different shapes

>>> M,N = 5,4
>>> q = get_neighbors(a.index('i'),(M,N))
2d index: (1, 3)
>>> q(a)
('c', 'd', 'e', 'j', 'o', 'n', 'm', 'h')
>>> M,N = 10,2
>>> q = get_neighbors(a.index('i'),(M,N))
2d index: (0, 8)
>>> q(a)
('j', 't', 's', 'r', 'h')
>>> M,N = 2,10
>>> q = get_neighbors(a.index('i'),(M,N))
2d index: (4, 0)
>>> q(a)
('g', 'h', 'j', 'l', 'k')
>>>

There is a nice discussion in the Numpy docs about making/treating a 1d thing as a Nd thing - Internal memory layout of an ndarray

The way you depicted your 1d --> 2d transformation used a column major scheme. I'm used to thinking row-major - I wrote the function to accept/expect a (nrows,ncols) shape argument but inside the function I kinda switched to column major processing. I was having to be careful so maybe that was a bad design.

How to determine neighboring elements from a 1D array?

1 Answers1