MATLAB Answers

Knut
0

Indexing irregular, constant width "stripes" of a 2-d array

Asked by Knut
on 18 Nov 2016
Latest activity Commented on by Knut
on 18 Nov 2016
I want to index an array in asynchronous stripes of constant width. Perhaps best described by an example:
bob = magic(7);
w = 2;
c = [1 3 6];
r = [1 4 3];
for stripe = 1:length(c)
res(stripe,:) = bob(r(stripe), c(stripe)+(0:w-1));
end
bob =
30 39 48 1 10 19 28
38 47 7 9 18 27 29
46 6 8 17 26 35 37
5 14 16 25 34 36 45
13 15 24 33 42 44 4
21 23 32 41 43 3 12
22 31 40 49 2 11 20
res =
30 39
16 25
35 37
Can the experts recommend a way to generally express this that is compact, readable and fast?

  1 Comment

Not clear, you want to extract elements with a common difference/ width? Check res, the difference is 9,9,2.

Sign in to comment.

Tags

Products

1 Answer

Answer by Walter Roberson
on 18 Nov 2016

Anything much different than that stops being as readable.
You could think in terms of calculating the linear indices of the sources using sub2ind. Once you have that then you can see how the entire source could be vectorized, and if the destination does not exist (or is being completely overwritten) you could construct the destination with reshape() instead of sub2ind for that.
You could then take the step of bringing the sub2ind inline. For 2d arrays it is ((row_number-1)*number_of_rows+column_number)
That would increase your speed. But it is less readable.

  1 Comment

Yes, I was thinking along those lines. Doing the sub2ind thing manually, exploiting the "outer sum" feature of recent MATLAB implicit expansion:
[ht,wd] = size(bob);
id2 = (c-1).*ht+r;
bob(id2+ht*(0:w-1)')'
one-liner (giving the transpose of the reference):
bob(((c-1)+(0:w-1)')*size(bob,1)+r)
But I agree, this code will be fast, but it:
1. Will be as comprehensible to general programmers as LISP is to me
2. Won't run on MATLAB 2016a or older

Sign in to comment.