efficient way of transforming vector

Hi,

I have a rather simple question on vector manipulation. I want to do some matrix manipulation as efficiently as possible and I think I found a solution, but therefore I have to do some vector manipulation. For example, I want to transform vector A into vector B:

A =

     2
     1
     0
     3

B =

     1
     1
     2
     4
     4
     4

So each element of A shows how many times the number 1,2,3,4... should be repeated. Is there a way to do this transformation in an efficient way without setting up a loop myself?

Thanks for your answer!

[Jos (10584)]

Just for your information, this is an example of run-length decoding. There are very efficient contributions available on the Matlab File Exchange for encoding and decoding run lengths (see, e.g., RUDE).

[the cyclist]

@Jos, good comment. My first thought on seeing this question is that I was pretty sure that it was covered in the classic "tips and tricks" document of Peter Acklam (still available on the web here: http://home.online.no/~pjacklam/matlab/doc/mtt/doc/mtt.pdf).

I was correct in my memory, but it turned out that the three algorithms he offers do not correctly handle cases where there is a zero run-length.

Answer 1

[Andrei Bobrov]

A =[2
     1
     0
     3];

n = A ~= 0;
k = cumsum(A);
k1 = k(n);
s = (1:numel(A))';
s1 = s(n);

k2 = k1 - A(n) + 1;
ii = zeros(k(end),1);
ii(k2) = 1;
idx = cumsum(ii);

out = s1(idx)

Answer 2

[Azzi Abdelmalek]

B=cell2mat(arrayfun(@(x) x*ones(A(x),1),1:numel(A),'un',0)')

%or a for loop which is much faster

B=[]
for k=1:numel(A)
  B(end+1:end+A(k),1)=k*ones(A(k),1)
end

Answer 3

[Lokesh Ravindranathan]

b is the resultant vector

b = [];
j = 1;
for i = 1:numel(a)
b(j: j + a(i) - 1) = i;
j = j+ a(i);
end