Multidimensional array indexing question

I have a matrix x that is of size [61 2 45].

linearIndex = find(x(:,1,:) < x(:,2,:));
xAverage = (x(:,1,:) + x(:,2,:))/2;

Now I want to assign the average to anywhere x(:,1,:) < x(:,2,:). I come up with the following but it seems a bit verbose and un-elegant. Thoughts on how to do this better?

[subScriptIndex1, subScriptIndex2, subScriptIndex3] = ind2sub(size(linearIndex), linearIndex);
x(subScriptIndex1, 1, subScriptIndex3) = xAverage(subScriptIndex1, 1, subScriptIndex3);
x(subScriptIndex1, 2, subScriptIndex3) = xAverage(subScriptIndex1, 1, subScriptIndex3);

[Jason Nicholson]

The new code looks like this:

linearIndex = linearIndex(:,[1,1],:);
x(linearIndex) = xAverage(linearIndex);

This is cleaner.

Answer 1

[Stephen23]

Your understanding is correct: if a logical index is shorter than the array it is being used on, then the index is not expanded in any way. The solution is to make the index the exact size that you require:

x = reshape((1:18)',[3 2 3])
xx = x;
idx = x(:,1,:) < x(:,2,:);
idx = idx(:,[1,1],:) % or repmat
x(idx) = nan
[xx(:) x(:)]

[Jason Nicholson]

Excellent. That works!

[Stephen23]

Note that this behavior is closely related:

>> X = [1,2,3,4];
>> X([false,true]) % shorter than X
ans =
     2
>> X([false,true,false(1,200)]) % longer than X, but only false..
ans =
     2

[Jason Nicholson]

Interesting.

Answer 2

[Fangjun Jiang]

x=rand(6,2,4);
MeanX=mean(x,2);
idx=x(:,1,:) < x(:,2,:);
x(idx)=MeanX(idx);

[Jason Nicholson]

The meanX line is better than what I did for sure.

The line starting with x(idx) doesn't work though. You don't get an error but it is wrong. The problem is the size of idx is [6 1 4] while x is of size [6 2 4]. I think what is happening is logical indexing relies on the linear index of an array. Therefore size of idx has 24 elements while x has 48 elements. Only the first 24 elements of x are modified while the 2nd 24 elements are not modified. Try the following to see what I am saying. Note that the 2nd half according to the linear index of x is not modified. The [xx(:) x(:)] when output to the command window, should make this clear.

x = reshape((1:18)',[3 2 3])
xx = x;
idx = x(:,1,:) < x(:,2,:);
x(idx) = nan
[xx(:) x(:)]

[Fangjun Jiang]

How about this? I think it works but what if the second dimension is larger than 2? There must be a a better way.

x=rand(6,2,4);
MeanX=mean(x,2);
MeanX(:,2,:)=MeanX(:,1,:);

idx=x(:,1,:) < x(:,2,:);
idx(:,2,:)=idx(:,1,:);

x(idx)=MeanX(idx);

[Jason Nicholson]

Your second suggestion does work.

When the dimension is greater than 2, I think repmat may be the solution.