How to pre-allocate for unknown size vector?

vector = [];
for i = 1:1000000
    r = randi([1 4],1,1); % a random integer generator between 1 and 4
    vector = [vector 1:r];
end

% 1. How to preallocate 'vector' ?
% 2. And get rid of unassigned variables ?
% 3. Preallocation considering Matlab coder compatability perspective !!
% 4. Can i use 'coder.nullcopy' ?

[Raymond MacNeil]

Good question!

Answer 1

[Image Analyst]

Here is one way:

numIterations = 4;
maxLength = 4;
vector = zeros(1, numIterations * maxLength);
lastIndex = 0;
for k = 1: numIterations
	r = randi([1 maxLength],1,1); % a random integer generator between 1 and 4
	vector(lastIndex+1 : lastIndex + r) = 1 : r;
	lastIndex = lastIndex + r;
end
% Crop off unused elements
vector = vector(1:lastIndex)

Answer 2

[John D'Errico]

There are several simple ways to do this.

1. Over-allocate the array. r will never be larger than 4. Therefore, the array will never be larger than 4e6 in length. So preallocate r to be of size 1x4e6. Now just index into the array, keeping a running count of where you stuffed the last elements into the array. At the end, just delete the unused elements. I would pre-allocate with NaNs, so you know what was unused.
2. Preallocate a cell array instead. You know how many passes through the loop. So create a cell array of size 1x1e6. Then stuff one cell at each iteration of the loop. At the end, just collapse the cell array into a flat array using cell2mat.

There are other ways to do this, but those are the easy ways. #2 is by far the better, and simpler, so use it. It will be fairly efficient, and very easy to read and debug. For example...

C = cell(1,1e6);
for i = 1:1000000
    r = randi([1 4],1,1); % a random integer generator between 1 and 4
    C{i} = 1:r;
end
vector = cell2mat(C);

whos vector
  Name        Size                    Bytes  Class     Attributes
  vector      1x2501228            20009824  double              

[JAI PRAKASH]

Thanks for your consideration.

I tried your suggestion#2.

Unfortunately cell2mat() is not supported for code generation. I don't know why, horzcat is also giving some error.

[John D'Errico]

Then better will be #1 I guess. It requires only that you keep acounter of where you are stuffing new elements into your array, then delete the unused portion of the array at the end. Extra bookkeeping that the cell array solution allows you to avoid.

I would point out that #1 is the solution that Image Analyst posed.

[JAI PRAKASH]

OK thank you very much

Answer 3

[Walter Roberson]

Start with an empty vector. Keep track of the number of positions used so far (which will start at 0)

At any point when you need to insert a 1:r, check to see if there are at least r positions remaining in the vector. If there are not at least r positions remaining in the vector, extend the vector by max() of r and (number of iterations remaining times average length of extension (that is, (4+1)/2)) . End if. Now insert the 1:r and increment the count of used values.

At end, trim out any unused locations.

vector = [];
vused = 0;
mrand = 4;
avgrand = (mrand+1)/2;
numIterations = 100000;
for i = 1 : numIterations
    Lvec = length(vector);
    r = randi(4);
    if Lvec-vused < r
        extend_to = Lvec + max(r, ceil((numIterations - i)*avgrand));
        fprintf('extending to length %d on iteration %d\n', extend_to, i);
        vector(extend_to) = 0;   %zero fills between current end and here
    end
    vector(vused+1:vused+r) = 1:r;
    vused = vused + r;
end
fprintf('trimming %d entries\n', length(vector)-vused);
vector(vused+1:end) = [];

If you test this, you will see that the number of extensions is quite low, often being only 2 or 3 times.