Using matrix index coming from max function to get values from another matrix

Dear All, I have the following problem:

A = randn(5,5);
B = randn(5,5);
C = randn(5,5);

D = cat(3,A,B,C);
E = randn(5,5,3);

[M,I] = max(D,[],3);

How to use the index matrix I to get respective elements from matrix E ? It means for every combination of i and j I want to take the value of E(i,j,:) whose index is given by the value of I(i,j).

I will be grateful for any comments!

Answer 1

[Star Strider]

This seems to work:

A = randn(5,5);
B = randn(5,5);
C = randn(5,5);

D = cat(3,A,B,C);
E = randn(5,5,3);

[M,I] = max(D,[],3);
IL = sub2ind(size(D), cumsum(ones(size(D,1)),1), cumsum(ones(size(D,2)),2), I);
Check = D(IL);                                      % Check To Be Sure Indexing Is Correct
Result = E(IL);

The ‘Check’ assignment demonstrates that the indexing reproduces ‘M’. It is not necessary for the code, and can be deleted.

[Adam Pigon]

Works perfectly! Thanks!

[Star Strider]

As always, my pleasure!

[Adam Pigon]

I have a small additional question : what if my problem takes more dimensions, i.e. matrices A,B,C are like size(A) = (5,5,5) ? How should I proceed then? I cannot fully understand the function sub2ind.

[Star Strider]

My code should work for all sizes of 3D matrices, since it uses the size function to create the relevant index matrices. (My code could be extended using the same idea to 4D and higher-dimension matrices. I have not explored that possibility.)

MATLAB actually addresses matrices with ‘linear indexing’, a segment of adjacent memory locations, not subscripts. The sub2ind function converts subscript indices to linear indices, that in many instances are the only effective way to index into a matrix, as with your matrix here. (The related ind2sub function does the reverse.)