Variable Indexing for N-dimension data

I am using Matlab R2013b. I've been struggling with Matlab indexing being a bottleneck in my functions. The issue keeps coming up. How can I tinker with Matlab's 'subsindex' function?

Overview of my problem: ------------------------------- I have a function that takes in a matrix of N-dimension. The function does not know the dimensions of the matrix beforehand. After some loops and things, the function must access data in the matrix. There's another function I have that creates a variable based on the number of inputs, essentially creating the same problem. I currently use 'sub2ind' to access or place values properly, but the 'sub2ind' function is slow. So slow it is a serious issue.

What I would Ideally like to do: ------------------------------------------ Let's say we have matrix A of N dimension. I have an index variable, indV. I want to be able to write A({indV}) without Matlab screaming an error at me about 'subsindex' not being defined for class 'cell'.

For example, if size(A) = (5, 4, 6) and indV=[1, 2, 3] then --> A({indV}) would be interpreted the same as A(1, 2, 3) resulting in one value. As you know matlab would produce three values if I wrote A(indV) which is seen as A(1:3). It would be a dream come true if I have the same A matrix and indV=[1 2] now and matlab would interpret A({indV},:) as A(1,2,:) resulting in 6 values instead of an error. What if indV1=3 and indV2=4 so A({indV1}, :, {indV2}) would be interpreted as A(3,:,4) producing 4 values, wouldn't that be nice. Even nicer would be with indV={2:4, 1} A({indV},:) would be seen as A(2:4, 1, :). You get the point.

I am hoping there is a way to modify/add to the 'subsindex' function to accomplish this resulting in a faster method than the 'sub2ind' function. Or maybe there is already functionality in Matlab that I am not aware of?

Any ideas/suggestions?

Answer 1

[Matt J]

Or maybe there is already functionality in Matlab that I am not aware of?

I think so. If you have indV={1,2,3}, then A(indV{:}) is equivalent to A(1,2,3). See Comma Separated Lists.

[Solagens]

Thank you. I knew there had to be something.

[Jan]

It works even with:

indV = {3, ':', 4}

[James Tursa]

Of course, using this directly works. For some reason I was fixated on the idea that calling subsref directly was needed when ':' was present, hence my overly complicated answer below!

Answer 2

[James Tursa]

The procedure for ':' functionality using the subsref ( ) function

indV = your cell array of indexing, which could include ':' entries
S.type = '()';
S.subs = indV;
result = subsref(A,S);

If you want to allow for indexing like {2,3} to behave like {2,3,':'} where ':' is automatically used for all unspecified trailing dimensions, then you could do this instead:

S.subs = [indV repmat({':'},1,numel(size(A))-numel(indV))];

E.g.,

>> A = reshape(1:24,2,3,4)
A(:,:,1) =
     1     3     5
     2     4     6
A(:,:,2) =
     7     9    11
     8    10    12
A(:,:,3) =
    13    15    17
    14    16    18
A(:,:,4) =
    19    21    23
    20    22    24
>> indV = {':',2:3,':'}
indV = 
    ':'    [1x2 double]    ':'
>> S.type = '()'
S = 
    type: '()'
>> S.subs = indV
S = 
    type: '()'
    subs: {':'  [2 3]  ':'}
>> result = subsref(A,S)
result(:,:,1) =
     3     5
     4     6
result(:,:,2) =
     9    11
    10    12
result(:,:,3) =
    15    17
    16    18
result(:,:,4) =
    21    23
    22    24

[Amr]

This is awsome. I have been using MATLAB for more than 10 years, and I did not know about this subsref() function. Thanks for sharing.