Divide a Large Matrix into Smaller Matrices in a Loop

Hi all,

I have a Matrix of dimensions 20x5.

I need to run calculations, using For loops, on a program that first needs the first 4 lines of the large matrix, then the next 4 lines etc...

In other terms: 1 iteration ---> Large matrix lines [1 to 4 x 5] .... used for the calculations. Then 2 iteration ---> Large matrix lines [5 to 8 x 5] .... used for 2nd iteration calculations. Then 3 iteration ---> Large matrix lines [9 to 12 x 5].... used for 3rd iteration calculations etc...

Does anyone has an idea on how to make it happen?

Answer 1

[Stephen23]

You can define the indexing quite easily:

mat = rand(20,5);
row = 4; % rows per group
grp = size(mat,1)/row;
for k = 1:grp
   idr = (1:row)+(k-1)*row;
   mat(idr,:)
end

Or split the data into a cell array using mat2cell:

C = mat2cell(mat,row*ones(1,grp),size(mat,2));
for k = 1:grp
    C{k}
end

[Nicolas Jalbert]

Stephen,

Thanks for your answer. I need to use the first one, since I need the matrices for the calculations.

One question remains: would it be possible to "store" this 5 news matrices created? ---> In the program I would then call one small matrix at a time via let's say: mat(1), then I call mat(2) ... This would render the program more simple since I could perform this matrix division before entering the calculations.

I would then only include in the calculations a for loop calling each of the small matrix at a time.

Would it be possible?

[Stephen23]

"Would it be possible?"

Yes. That is exactly what the second method shows you how to do: C is a cell array containing all of the submatrices. Access them using cell array indexing:

C{1}
C{2}
etc

Or, if you really want to use indexing rather than mat2cell, then just put them into a cell array explicitly:

C = cell(1,grp);
for k = 1:grp
    idr = (1:row)+(k-1)*row;
    C{k} = mat(idr,:);
end

[Nicolas Jalbert]

I found a way without having to use cells arrays, this way we keep the matrices up for use. Because if they are stored in a cell array, we need a conversion (cell2mat?) to perform calculations with them right?

Here is my program:

mat = rand(20,5)

row = 4 % rows per group

grp = size(mat,1)/row

for k = 1:grp

idr = (1:row)+(k-1)*row;

mat(idr,:);

Matrix(:,:,k) = mat(idr,:);

end

Matrix

Is this a proper way to store these matrices? Or is there more elegant?

[Stephen23]

"Because if they are stored in a cell array, we need a conversion (cell2mat?) to perform calculations with them right?"

Not at all. You just need to use indexing, or something like cellfun.

"Is this a proper way to store these matrices?"

What is your definition of "proper"? The best way to store data is determined by the nature of that data and how is going to be processed, neither of what you have said anything about. Keeping that data in one numeric array will probably make numeric calculation a lot simpler, as you could write vectorized code.

"Or is there more elegant?"

An elegant way to rearrange that matrix into a 3D array is to use reshape and permute, no loop is required:

out = permute(reshape(mat.',[],row,grp),[2,1,3])

Answer 2

[Andrei Bobrov]

mat1 - matrix with size (20,5)

row1 = 4;
Matrix1 = permute(reshape(mat1.',size(mat1,2),row1,[]),[2,1,3]);