Elementwise indexing of a vector by a multidimentional array

I have:

• A, an n-dimentional array of integers, whose values are in the range 1-100,
• and T, a 1-dimentional, length 100 array of values.

I want a new n-dimentional array B which, for each corresponding (same position) integer c of A, has the value T(c).

What is the proper way to do this without a for loop, preferrably that works on a gpuArray?

Answer 1

[KSSV]

n = 10 ; 
A = randi(n,2,2,10) ;
B = rand(1,n) ;
iwant = B(A)

iwant = 
iwant(:,:,1) =

    0.6011    0.4264
    0.6566    0.7867


iwant(:,:,2) =

    0.0678    0.6011
    0.2322    0.6011


iwant(:,:,3) =

    0.0678    0.4264
    0.2878    0.2322


iwant(:,:,4) =

    0.9453    0.2322
    0.6598    0.0678


iwant(:,:,5) =

    0.7867    0.6598
    0.9453    0.7867


iwant(:,:,6) =

    0.4264    0.0678
    0.5672    0.2322


iwant(:,:,7) =

    0.2878    0.9453
    0.0678    0.9453


iwant(:,:,8) =

    0.9453    0.9453
    0.6566    0.6566


iwant(:,:,9) =

    0.6598    0.4264
    0.6566    0.6566


iwant(:,:,10) =

    0.6566    0.5672
    0.5672    0.2878

Answer 2

[DGM]

I have no idea what's ideal for gpuArray, but this is how it would be done with linear indexing.

T = 11:20; % a vector
A = randi([1 10],2,2,2) % a 3D array

A = 
A(:,:,1) =

     3     8
     5     2


A(:,:,2) =

    10     3
     6     9

B = T(A) % linear indexing

B = 
B(:,:,1) =

    13    18
    15    12


B(:,:,2) =

    20    13
    16    19