Vectorization fail with sub2ind?

I have the for loop code
for m=1:numel(N,1)
y = N(m,1); % y matrix index
x = N(m,2); % x matrix index
dy = N(m,4); % codfficient
dx = N(m,3); % coefficient
wts(y,x) = wts(y,x)+dx.*dy;
end
I attempted to vectorize this as follows
ind = sub2ind(size(wts),N(:,2),N(:,1));
wts(ind) = wts(ind)+N(:,3).*N(:,4);
Unfortunately, for some reason the result is totally wrong.
Someone later said that repeated values of "(x,y) / ind" will be erroneous. I don't know what that means, but there are many repeated values of the sub2ind index in my code (lots of addition).
How can this be fixed?

4 Comments

The documentation, numel, Number of array elements, says that numel takes one argument. However,
>> numel( magic(6),1 )
ans =
1
did you intend
>> size( magic(6),1 )
ans =
6
What is the initial value of wts?
What do you mean wrong result? The vectorized code looks right to me.
per isakson
per isakson on 5 Mar 2014
Edited: per isakson on 5 Mar 2014
Why not show an executable example based on a for-loop?
I have built the following script which I think is illustrative:
clear all
xygrd=10; % grid height, width
wts=zeros(xygrd,xygrd); % preallocate
wts2=zeros(xygrd,xygrd); % preallocate
% create points randomly allocated to the matrix xygrd*xygrd by randomly
% picking x,y indices
N1 = floor(rand(1000,2)*xygrd)+1; % x,y indices
N2 = rand(1000,2); % point values (dx,dy)
N=[N1 N2]; % x,y,dx,dy
for m=1:size(N,1)
y = N(m,1); % y matrix index
x = N(m,2); % x matrix index
dy = N(m,4); % dy
dx = N(m,3); % dx
wts(y,x) = wts(y,x)+dx.*dy;
end
% vectorization
ind = sub2ind(size(wts2),N(:,2),N(:,1));
wts2(ind) = wts2(ind)+N(:,3).*N(:,4);
pcolor(wts-wts2); % should all be zeros
matrix wts (from the for loop) is completely different from wts2. I need a vectorized code for the for loop.
Thanks for thinking about this with me.

Sign in to comment.

 Accepted Answer

Roger Stafford
Roger Stafford on 5 Mar 2014
Edited: Roger Stafford on 5 Mar 2014
Here's a simple example to show you how repeated values of 'ind' will produce different results from your for-loop:
wts = zeros(3,1);
ind = [2;2;2];
wts(ind) = wts(ind) + [3;7;19];
wts
ans =
0
19
0
In your vectorization you would have been expecting [0;29;0] as in a for-loop, but instead only the last addition is performed, the other two results having been overwritten.
To do what you want, use matlab's 'accumarray' function:
wts = wts + accumarray([N(:,1),N(:,2)],N(:,3).*N(:,4),size(wts));
It will perform all the additions, even with repetitions.
(Note: You appear to have N(:,2) and N(:,1) in reversed order in your call to 'sub2ind' which would produce further errors.)

1 Comment

Thanks Roger, It appears you have found the correct method!
% diff vectorization
wts3 = wts3 + accumarray([N(:,1),N(:,2)],N(:,3).*N(:,4),size(wts));
is can be added to the script posted earlier, and it reproduces the for-loop.

Sign in to comment.

More Answers (0)

Categories

Find more on Loops and Conditional Statements in Help Center and File Exchange

Products

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!