Vectorized generation of a cell

x = 1 : a; y = 1 : b; z(x,y) = {randperm(d,2)}; a non-repeating cell is desired. No for loops please.

[Jon]

How are the values of a and b used. What is the value of d?

[Amir Mahmoudi]

a, b, d are any numbers.

Answer 1

[Matt J]

There is no way to create or manipulate cells without for-loops. Things like mat2cell, num2cell, etc... all have for-loops inside them. So do things like cellfun, arrayfun, etc...

It is not clear why you would want a cell array. Your data can be stored in an a-by-b-by-2 array.

a=5; b=3; d=7;

I=nchoosek(1:d,2);
I=[I;fliplr(I)];
r=randi(height(I),a*b,1);

z= reshape(I(r,:)   ,a,b,2)

z = 
z(:,:,1) =

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


z(:,:,2) =

     7     5     5
     5     3     4
     5     7     7
     6     7     6
     2     3     6

[Amir Mahmoudi]

not even with cellfun?

[Amir Mahmoudi]

In optimization processes (with large numbers), vectorized forms work faster than for loops.

[Matt J]

not even with cellfun?

No, cellfun is no faster than a for-loop. It is just a way to hide the for-loop inside a 1-line piece of code.

Again, though, there is no reason to be holding your result in a cell. Did you try my suggested code?

[Dyuman Joshi]

Like mat2cell() and num2cell() mentioned by Matt, cellfun() is a for loop in disguise as well.

A simple for loop would be a much better choice.

"In optimization processes (with large numbers), vectorized forms work faster than for loops."

Yes, if properly utilized, vectorized code is generally faster than for loops. However, that applies to numeric arrays, not cell arrays.

It will be better to work with numeric arrays, as Matt has suggested and shown above.

[Amir Mahmoudi]

Thanks for your response. In fact, this is just a small part of a large coding. It must be a cell. If there is no way, then I have to use for loops.

[Dyuman Joshi]

Do you have a constraint to only use a cell array? If yes, then what is the constraint?

[Amir Mahmoudi]

Thanks for useful comments.

[Matt J]

You're welcome, but if the answer resolves your question, please Accept-click it.

[Bruno Luong]

r=randi(height(I),a*b,1);

Shouldn't be

r = randperm(height(I),a*b);

if you want all "cells" have different pairs?

[Amir Mahmoudi]

I accepted the answer because of the comments. The answer is not what I wanted.

[Matt J]

It must be a cell. If there is no way, then I have to use for loops.

If you don't care about avoiding for-loops anymore, you can take the result of my proposed code and convert it to a cell with num2cell:

a=5; b=3; d=7;

I=nchoosek(1:d,2);
I=[I;fliplr(I)];
r=randi(height(I),a*b,1);

z= reshape(  num2cell(I(r,:),2)   ,a,b)

z = 5×3 cell array
    {[6 2]}    {[2 1]}    {[5 7]}
    {[4 5]}    {[3 5]}    {[5 7]}
    {[1 2]}    {[7 3]}    {[4 7]}
    {[7 5]}    {[3 7]}    {[1 4]}
    {[1 4]}    {[2 4]}    {[2 6]}