About preallocating for speed

Dear Contributers,

There is a " for loop" in my program and Matlab gives me a suggestion to consider "Preallocating" for speed. I want to learn and do that. However, I don't know where to start to learn it. Of course, I looked the documents in mathworks but I found a few documents in there. I can read all of them but I don't want to rush headlong into this topic. Please, I am asking your ideas where I should start from those documents;

http://www.mathworks.com/help/matlab/creating_plots/preallocate-arrays.html?searchHighlight=preallocate

http://www.mathworks.com/help/matlab/matlab_prog/preallocate-memory-for-a-cell-array.html?searchHighlight=preallocate

http://www.mathworks.com/help/matlab/math/resizing-and-reshaping-matrices.html#f1-88760

By the way, My code is;

for k1 =1: length(X)  
    for k2 =1: length(X_inv)
            DE(k1,k2)= hypot(X(k1)-X_inv(k2), Y(k1)-Y_inv(k2));
    end
 end

And Matlab suggested me to preallocate the DE

I am not sure whether this question is appropriate for here or not. If it is not okay, accept my apologies.

Answer 1

[John D'Errico]

When you grow an array incrementally as you are doing with DE, MATLAB is forced to reallocate the entire array at EVERY iteration, copying over the entire mess just to add ONE element. This will cause the operation to get slower, and the time required will grow quadratically. SO it will get SLOW.

You know in the end EXACTLY how large DE will be. So just add ONE extra line before the loop.

DE = zeros(length(X),length(X_inv));

This essentially creates the array in advance, filling it with zeros. It need no longer reallocated at each iteration, because it will no longer need to change size at every step.

Much of the time, preallocation is not needed. MATLAB is not a language where you need to initialize/allocate every array. Variables that are scalars, and will remain so are never an issue. It is only when an array is dynamically grown that preallocation is important. A problem arises when you don't know the final size of your array. Even there, there are tricks that one can do. I posted a submission to the FEX long ago that allows the user to store information in a form that can be more efficiently grown, then at the end, you can unpack the object into a regular array.

[Ender Rencuzogullari]

Program is ,now, working extremely fast. Marvellous! Thank you. I want to learn the logic of that much more now. I will read that file as soon as possible soon. Thank you again, sir.

[John D'Errico]

I'm always happy to see & help someone wanting to learn, as opposed to see someone wanting their homework done for them. :)

[SINDU GOKULAPATI]

hey john im facing a similar issue , for context below is my code

a=csvread('D:\sem6\PROJECT\hygdata.csv',1,5,[1,5,5068,7]);  %reading x,y,z
n=5068;
//r = zeros(1,n) %error:Unable to perform assignment because brace indexing is not supported for variables of this type.
for i=1:n
    r{i} = transpose(a(i,:));  %mztrix [x y z]
end

what is the alternative here?

[Stephen23]

"what is the alternative here?"

Preallocate the array using the correct class:

r = cell(1,n)

Answer 2

[Madison Schossow]

try doing length(X):-1:1