double indexing

Hello,

is it possible to do double or multiple indexing in matlab? E.g.

>> v = [1,2,5,7,9]
v =
   1     2     5     7     9
>> v(1:4)
ans =
   1     2     5     7
>> v(1:4)(2:3)
ans =
   2     5

The first index create a new vector and the second index creates a new vector out of the newly created one. It is common in other languages, and it helps to avoid defining temp variables.

Cheers, Frank

[Katharina Wellstein]

A simple way to do this would be by using intersect, a function that finds commonalities between different arrays.

In this case it would be:

[value, position] = intersect(v(1:4), v(2:3))

For more complicated things like indexing into an array (or matrix) with different indices, the same thing can be used. E.g. if I want to find numbers that are higher than 5 and numbers that are even in x, I can do the following:

x = randi(randi(10,10,1);
bigIdx = find(x>5);
evenIdx = ~mod(x,2);
x(intersect(bigIdx,evenIdx);

[Walter Roberson]

The hypothetical v(1:4)(2:3) is intended to select elements 1 through 4 of v, and then select elements 2 to 3 of that.

Suppose we had instead asked for v(2:4)(2:3), then that would be elements 2 to 3 out of elements 2 to 4, so the end result would be v(3:4) .

[value, position] = intersect(v(2:4), v(2:3))

would give back v(2:3) instead, at best.

If we had

v = [1,5,5,7,9]

then [value, position] = intersect(v(1:4), v(2:3)) would be interesect([1 5 5 7], [5 5]) which is going to return [5] rather than [5 5] because the return values of intersect() are sorted and unique by default.

Overall, intersect() is just the wrong thing to use here.

Answer 1

[Walter Roberson]

The closest you can get is

subsref(V(1:4), struct('type', '()', 'subs', 2:3)))

You can follow {} referencing by () referencing, {}(), but you cannot use ()() or (){} or ().{} or ().(), and you cannot use function(){} or function()() or function().field

[Frank]

thanks. Then I prefer to use a temp var instead. :)

[Petter]

"You can follow {} referencing by () referencing, {}()"

This one line made my day so much easier, greatly appreciated. Thumbs up, gold star.

[Walter Roberson]

In the time since the above Answer was posted, Mathworks has since permitted function().field

Answer 2

[mklcst mklcst]

I think it could be very useful to have a short way to perform double indexing.

Answer 3

[Jos (10584)]

If the indices are stored in variables, this is trivial!

V = [1,2,5,7,9]
ii = 1:4
jj = [2 3]
out = V(ii(jj))

[lee eugene]

However, if the two index do not have the same starting indexing, there would be something wrong. For example, i would like to select index [2,4] from [1,2,5,7,9] at first, and then select index [1,2]. Then it would be [2,5,7] at first and [2,5] in the end.

[DGM]

That's what the result is supposed to be.

That said, there's still room for problems if the index vectors are programmatically generated. We'd have to safeguard against out-of-range indices. Consider one possibility:

% a vector
V = [1,2,5,7,9];

% potentially out-of-range indices
ii = [1:4 7];
jj = [2 3 5];

% one option might be to simply 
% discard invalid indices
ii = intersect(ii,1:numel(V))

ii = 1×4

1 2 3 4

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

jj = intersect(jj,1:numel(ii))

jj = 1×2

2 3

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

% so it still works without indexing errors
out = V(ii(jj))

out = 1×2

2 5

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

I'm sure there are other ways one might choose to handle it though. This might not be okay if you're not expecting to a different number of elements than what you requested. In that case, it might be better to just throw an error.