Is there another way to use mat2cell more efficiently when facing large matrices?

My project right now need to handle extreme size matrices. But I need to split them into equal sizes matrices in terms of a cell column for further calculation.

Right now my method of approach:

Assume A1 is the targeted matrix, where A1= 192x8

If I need to split it into each 8x8 matrix, well 192/8 = 24 cells.

My working on this matter:

d = mat2cell(A1U,[N N N N N N N N N N N N N N N N N N N N N N N N],N) %where N = 8

d = 24x1 cell group

{8×8 double}

{8×8 double}

{8×8 double} ..... total of 24

Although right now I can still handle these kind of sizes, but my target was a 1474560 x 64 matrix.

If so then I need to type 1474560 / 64 = 23040 times of N in the equation.

Is there a better way to tackle this problem?

[Moses Tang]

Or should I choose another way to split it into cell column formats?

Maybe consider each sets of rows using a for loop?

But then need to reform it into a cell column...

Not sure what would be the better approach...

Answer 1

[madhan ravi]

Use a ND array instead of cell array:

permute(reshape(A1.',size(A1,2),[],size(A1,1)/N),[2 1 3]) % A - your matrix ,N = 8

[madhan ravi]

If you insist mat2cell() lookup the below commands:

help repmat 
%or
help repelem
% % to copy values of N n times

[Moses Tang]

it seems correct with multiple a+bi format matrices.

But I don't know how to read the line you just gave to me.

Could you teach me how to read to understand better how does it implement?

Is it from the page: https://uk.mathworks.com/help/matlab/math/multidimensional-arrays.html

[madhan ravi]

mat2cell(A1, repelem(N,24), N)

https://www.mathworks.com/help/matlab/ref/repelem.html

https://www.mathworks.com/help/matlab/ref/repmat.html?searchHighlight=repmat&s_tid=doc_srchtitle

[Moses Tang]

Alright thank you very much.

I will take a good look of it. Thanks for everything. ^^

Answer 2

[Guillaume]

If so then I need to type 1474560 / 64 = 23040 times of N in the equation.

Of course not! And you never had to type N more than once:

N = 8
numblocks = 64
d = mat2cell(A, repelem(N, numblocks), N)

Note that if you have the image processing toolbox, you could use blockproc to do the splitting and looping for you.

It may also be that you don't need to use a cell array at all (there's a fair bit of memory and speed overhead with cell arrays). Madhan's suggestion of using ND arrays is one way, or just looping over the indices in steps of 8.

[Moses Tang]

Thank you very much!

I will have a look on the method uing replelem.