How to use vectors instead of loops?

Hi, I am new to coding and I seem to be using too many loops/cells which significantly slows down my codes.

I want to define a 200x1 matrix every cell of which is a 8X17 matrix and the cells of this one are 1x3 vectors. The values in the 1x3 vectors are generated randomly (from different sets of data: Hmin, Hmax, and Wmax are 8x17 matrices of data which have been imported as cell arrays. They are simple numbers which provide a range for the randomly generated values in 1x3 vectors.).

I want to run a genetic algorithm and this would define 200 members of my initial populations. I have tried this in MATLAB using cell arrays; but they significantly slow down my code. This is what I have tried in MATLAB.

I would be grateful if someone would help me to avoid this.

P = cell (200, 1);
for i = 1:200
    P{i,1} = cell(8, 17);
    for j = 1:8
        for k = 1:17
            P{i,1}{j,k} = zeros(1,3);
            P{i,1}{j,k}(1,1) = randi(Hmax{j}(k));
            P{i,1}{j,k}(1,2) = randi([Hmin{j}(k),Hmax{j}(k)]);
            P{i,1}{j,k}(1,3) = randi(Wmax{j}(k));
        end
    end
end

Thank you so much.

[Matt J]

You are also using cell arrays over-liberally, for variables that could easily be normal numeric arrays. This, aside from the loops, will also slow you down considerably. Any recommended rewrite would need to know what Hmax, Hmin, and Wmax contain and their sizes. Could we just pre-convert them to LxD matrices with,

Hmin=cell2mat(Hmin)
Hmax=cell2mat(Hmax)
Wax=cell2mat(Wmax);

[dpb]

Also just a description of the problem trying to solve -- it's possible someone might see a more direct algorithmic change if knew what is intended instead of trying to trace through totally opaque uncommented code with no idea what the end result is supposed to be.

[Sherwin]

Thank you so much for mentioning that. Hmin, Hmax and Wmax are simple LxD matrixes like this:

2	1	1	1	2	2	2	2	2	2	2	1	1	2	2	2	3
2	1	1	2	2	3	3	2	2	1	1	2	2	2	3	3	3
1	1	2	2	3	3	3	2	3	2	3	3	2	2	3	2	5
2	2	2	2	2	2	3	3	2	2	2	2	2	2	3	3	4
3	1	1	1	3	3	3	3	4	4	4	3	2	2	5	5	5
3	2	2	3	4	4	4	4	4	4	4	4	5	5	5	5	7
3	2	2	2	2	2	2	2	3	3	2	2	2	2	2	3	3
2	1	1	2	2	2	2	2	2	2	2	2	2	2	2	2	2

However, I have imported them from excel as 8x1 (Lx1) cells (each row) seperately.

[Image Analyst]

Sherwin, Matt is exactly right when he said it's the cell arrays that are slowing it down, not the for loops. You have a tiny 27,200 iterations. This is nothing - goes faster than a greased cheetah on steroids. Look:

P = 200;
L = 8;
D = 17;
% Code 1
tic
for i = 1:P
	for j = 1:L
		for k = 1:D
		end
	end
end
toc

Elapsed time is 0.000066 seconds.

So the for loop part of it is only 66 microseconds. But cell arrays are very inefficient. Like Matt said, you could likely use regular numeric arrays for that. But like pdb said the code is an uncommented, alphabet soup mess of a program so we don't really see what it's doing so we don't know if cell arrays are really needed. (I'd never accept this by someone who works for me.) Plus we can't even run it because you forgot to attach or assign everything that is needed to run it (and I tried).

https://www.mathworks.com/matlabcentral/answers/6200-tutorial-how-to-ask-a-question-on-answers-and-get-a-fast-answer

Please see the FAQ for a good explanation of what cell arrays are, and how and when to use them.

FAQ: What is a cell array?

[Sherwin]

I highly appreciate everyone's help.

The first code is supposed to generate an array of (200x1) every cell of which is an array of 8x17 and the cells of this one are 1x3 vectors. The values in the 1x3 vector are generated randomly. Is it possible to use matrices/arrays instead of cell arrays in this case?

I guess if I figure this out, I would be able t imprve the rest of the code too.

[Walter Roberson]

Unless you are required to use cell arrays, you could do the same work with a 200 x 8 x 17 x 3 matrix.

[Sherwin]

Thank you so much.

I am not required but I find it easier to use cells which significantly slows down my code. So I want to learn how to that with arrays/matrices. Would you please explain it in code.

[dpb]

"... generate an array of (200x1) every cell of which is an array of 8x17 and the cells of this one are 1x3 vectors"

The question evolves to "Why is this a useful storage scheme?"

What is the end result to be used for that some other less convoluted scheme wouldn't work just as well, or better?

The beginning portion of M 2D arrays is handled pretty efficiently by a 3D array where each plane is a slice.

The next thing about a 3-vector at each of those elements would require something more, indeed, but not knowing the point, it's not certain what to do otherwise.

What about just generating the vector on the fly when it's needed? Or, for that matter any of the values if they're just randomized realizations.

[Sherwin]

Thank you for your comment.

Actually these are 200 members of my genetic algorithm population and need to be generated beforhand and then go through somefunctions (mutation, crossover, etc).

If you think that I can use a multidimentional array with 200 pages and each page a 8x17 matrix consisting of 1x3 vectors is a better option, would you please explain, how I should do that in code and how I should reference a cell in that kind of data storage method? (since when I use [ ] in matlab I get an error)

[David Goodmanson]

Hi Sherwin,

You have four indices all together: two for the 8x17 matrix, one for the 3 components of the vector, and one for the 200 members. You can store all of this stuff in a four-dimensional array. There are several ways to do that, but suppose the 8x17 matrix is seen as basic. Then a 4d array can be initialized as

P = zeros(8,17,3,200).

After the array is filled, then x = P(:,:,2,123) is an 8x17 matrix containing the second vector component, for member 123. x = P(:,:,:,123) is an 8x17x3 array that contains all the information about member 123. If you display it, it can be seen as three 8x17 matrices stacked in layers, each layer containing one of the components of v. P(5,9,1,:) is a vector that contains the (5,9) element of the matrix for vector component 1, listed out for all 200 members. And so on.

A slight complication is that operations like this give an object that is still considered to be a 4d array, but some of those four dimensions often have only one component. For example,

size(P(:,:,2,123))
ans = 8  17  1   1

You can use the squeeze operator to reduce the dimensionality.

m = squeeze(P(:,:,2,123))

results in a 2d matrix, size 8x17.

[Sherwin]

Thank you so much.

[Stephen23]

See also:

https://www.mathworks.com/help/matlab/math/multidimensional-arrays.html