Use NxD array of indices to reference a D dimension array in matlab

I have a matlab matrix full of indices (indx) which can be N rows and D columns (and varies from execution to execution). I also have an output array of D dimensions, output(D1,D2,..,D_D). For a given indx, for example if D=3, a row of indx might equal (1, 2, 3), and I'd like to index output(1,2,3)=k.

However, I am not sure of two things.

1. I don't know D beforehand, how can I access a D length set of indices in output?

2. If so, is there any way accessing all ordered n-tuples in indx for each of their associated entry in output?

For 1, a remark here seems relevant: mathworks tip, specifically the part about "Indexing into an N-D array with N subscripts without knowing N ahead of time (see fftshift for an example)", except a) I'm indexing into an ND array without knowing D, and b) looking at fftshift did not help.

To be completely clear, here's some code illustrating the problem

qq(:,:,1)=[1 2; 3 4];
qq(:,:,2)=[5 6; 7 8];
indx=[1 1 1; 2 2 2] %to reference the coordinates (1,1,1)=1, and (2,2,2)=8

Thanks for looking at my question.

Edit* Sorry about qq missing the '8'.

[Matt J]

Your qq(:,:,2) does not have enough entries and produces an error message. Did you mean

 qq(:,:,1)=[1 2; 3 4];
 qq(:,:,2)=[5 6; 7 8];

If so, then qq(2,2,2) would not be 7, not 8.

Answer 1

[Matt J]

For your specific example

 indxcell = num2cell(indx,1);
 lindx = sub2ind(size(qq),indxcell{:});

 result = qq(lindx)

[Matt J]

It is suspicious, however, that you are in the situation that you describe. When processing arrays of dimension greater than 2, the coordinates you want to reference are nearly always the result of some logical operation, e.g., all coordinates where qq<10.

In these cases, you never work with D-tuples of coordinates. You just use logical indexing

indx=(qq<10);
result = qq(indx);

Or, if you really need 'indx' in a compressed form, you would use linear indices,

 indx=find(qq<10);
 result=qq(indx);

[M S]

I see, you use linear indexing rather than dimensional indexing.

Working on a probability density problem in arbitrary dimensions, and I'd like to keep track of what neighboring coordinate positions are doing. It could be done with a list of positions, but finding neighbors is more cumbersome in list form vs. an DxDx..xD array. Thanks for your help!

[M S]

Matt -- any thoughts on alternatives? It seems this approach is very very inefficient. 22% of computational time of my program is spent on num2cell, and another 10% on sub2ind, and a further 10% on the actual reference. That seems ridiculous when I know the indices. Any way to avoid this?

One thing that should work would be to build indx as a cell array from the beginning, that would cut out num2cell...