Vectorizing the creation of an offset vector

Is there any possibility to vectorize this piece of code:

BIN_1 = 7;
C = randi(10,1000,1);
offset_vec = zeros(size(C));
delete = [];
o = 0;
for i = 1:size(C)
    if C(i) == BIN_1
        o = o + 1;
        delete = [delete, i];
    end
    offset_vec(i) = o;
end

It creates an offset vector, which increases whenever the corresponding element in C equals 7. The for-loop seem very slow.

Answer 1

[Andrei Bobrov]

ii = find(C == BIN_1);
offset_vec = zeros(size(C));
offset_vec(ii) = 1;
offset_vec = cumsum(offset_vec);

or (ADD)

t = C == BIN_1;
offset_vec = cumsum(t);
ii = find(t); % analog of 'delete'

[Jan]

Slightly faster: Omit the find() and use logical indexing.

[Andrei Bobrov]

Hi Jan! I agree with you. Added.

Answer 2

[Jan]

Some comments:

1. Letting a vector grow iteratively is a severe waste of time. Note, that if delete is a 1x1000 vector finally, the iterative construction demands for reverving sum(1:1000) elements any copying almost the same number of elements. Better:

delete = false(1, numel(C));
o = 0;
for i = 1:numel(C)
  if C(i) == BIN_1
      delete(i) = true;
  ...

2. size(C) replies a vector. Therefore for i = 1:size(C) might perform unexpected things. Better use (as shown in the code above already) numel(C) or size(C, 1) (or what ever you need exactly).

[Matthias]

I know that growing Arrays are pretty slow, but the Finale size of the Array is somewhere around 100, while the for loop has several million iterations. That's why I believe it doesn't really matter in this specific case.