Storing vectors of different length in one single vector

Hello, hope you can help me out here...

I have a for-loop which produces with every loop a vector of values. The vectors it produces can have different lengths (for example, the vector in the first round has 10 values, in the second round 15, in the third 2...). In the end I'd like to store the values of these vectors in one single vector, but i really don't know how to best do it.

Thanks a lot in advance!

Answer 1

[Jan]

The slow way:

x = [];
for i = 1:100
  y = rand(1, floor(rand*11));
  x = [x, y];  % or x = cat(2, x, y)
end

Now the vector x grows in every iteration. Therefore memory for a new vector must be allocated and teh old values are copied.

Filling a vector, cleanup afterwards:

x = nan(100, 10);
for i = 1:100
  n = floor(rand*11);
  y = rand(1, n);
  x(i, 1:n) = y;
end
x(isnan(x)) = [];  % Now x is a column vector

Collect the vectors in a cell at first:

C = cell(1, 100);
for i = 1:100
  C{i} = rand(1, floor(rand*11));
end
x = cat(2, C{:});

The overhead for creating the CELL is much smaller than the slow method of the growing array.

And if you the array is very large and the section is time-consuming, it matters that cat seems to have problem with a proper pre-allocation internally. Then FEX: Cell2Vec is more efficient.

Answer 2

[Shiva Arun Kumar]

Consider using cell arrays: http://www.mathworks.com/help/techdoc/matlab_prog/br04bw6-98.html

Answer 3

[Walter Roberson]

The problem isn't so much storing the variable-length information: the problem is in getting the information out afterwards.

You need to store one of:

• the location of the start of each section
• the length of each section
• a unique marker between sections that cannot occur in the data itself

If you choose to store length information, you can store it separately, or you can prefix each segment with the segment length.

For example:

vector_of_data = [];
for k = 1 : whatever
  newdata = as appropriate
  vector_of_data = [vector_of_data length(newdata) newdata];
end