Combine each row of matrix into a single vector within a cell array

Hello,

I have a big mxn matrix with m>100,000 and n=1800. Each row of this matrix contains measurement values for a specific point in time. For further processing of the data, I would like to create a mx1 cell array containing the rows of the matrix. Is there a possibility achieving this without a loop?

Here is my code, using a for loop:

A = rand(100000,1800);
b = cell(size(A,1), 1);

for ii = 1:size(b,1)
    b{ii,1} = A(ii,1:end);
end

Answer 1

[Jan]

This sounds like a job for num2cell.

A = rand(100000,1800);

tic
b = cell(size(A,1), 1);
for ii = 1:size(b,1)
    b{ii,1} = A(ii,1:end);
end
toc

Replace "1:end" by the faster ":" :

clear b  
tic
b = cell(size(A,1), 1);
for ii = 1:size(b,1)
    b{ii,1} = A(ii, :);
end
toc

Compare with num2cell:

clear b
tic
b = num2cell(A, 2);
toc

Even faster without the overhead of num2cell, but operating in columnwise order:

tic
A = A.';
b = cell(size(A,2), 1);
for ii = 1:size(b,1)
    b{ii,1} = A(:, ii).';
end
toc

Timings under R2016b, i7, 8GB RAM:

 Elapsed time is 3.165280 seconds.   Original
 Elapsed time is 2.672808 seconds.   1:end -> :
 Elapsed time is 2.386299 seconds.   num2cell
 Elapsed time is 1.717735 seconds.   columnwise processing

Answer 2

[Sujit Muduli]

Hello Tillmann,

I didn't find any such operation to fill the cell array in one shot without a for loop. I also don't know your exact use case but what I could suggest here is that you don't need to convert it into a cell array. But you can always get a row element of the matrix on demand, by this you may avoid a redundant for loop operation.

But if you could explain your use case and workflow briefly I may suggest you something else.

Thanks

Sujit