Using cell array to create 3D sparse matrix.

Hello, I would like to create a 3d-sparse matrix. Ofc, I know that matlab's toolbox doesn't have it so far.

Here is my code :

C=zeros(m^nun,m,nu);

I would like to use a cell array filled with 2d sparse matrix. Something like :

A = zeros(m^nun, 1);
B = cell(size(A));
SP = sparse(m, nu);

for j=1:m^nun
B{j} = SP;
end

Is it the right way to do it ?

Furthemore, can I use my matrix B in the same way as my matrix C ? For example, here is my code (and it works) :

C(1:x,A,1)=N;

But is it going to work with my matrix B as well ? (by doing : B(1:x,A,1)=N;)

Thank you, and sorry for the long post.

Answer 1

[Matt J]

Consider using this FEX contribution instead

http://www.mathworks.com/matlabcentral/fileexchange/29832-n-dimensional-sparse-arrays

It will let you do many/most things that you can do with 3D full arrays.

[Matt J]

Collet Bastien commented:

"Thank you very much, this contribution is rather useful. However, I'm still having a problem on :

prod(AP(:,A(nl+i,:)~=0,i),2); (where AP is a ndsparse object)

it's because prod is not defined for the objects ndsparse (??? Undefined function or method 'prod' for input arguments of type 'ndSparse'.).

What can I do to fix this problem ? I didn't find a documentation for this toolbox (this toolbox is quiet big, maybe the function I'm looking for already exists)."

[Matt J]

It looks like you are taking the product along a very small piece of AP, so as a workaround, you could simply do this:

prod(full(AP(:,A(nl+i,:)~=0,i)),2);

It's true that I didn't implement a PROD method for the class. I could put it on my TODO list if you really think it will be useful.

I didn't expect people to have much use for a PROD method because very often PROD(A,DIM) for a sparse array would give an array of all zeros. Also, even in cases where A is non-sparse along certain DIM, the dimensions of a sparse A are usually pretty large in practice, so PROD might often overflow in that case.

[Collet Bastien]

The prod method for a sparse matrix is only useful in some special cases (as you said, the most of the time, we will get 0 or an overflow). Even if I need it, I don't think it's very interesting to add this method to the toolbox.

I'm in trouble with my 3d matrix, maybe can you help me. I have this matrix :

A = (1,1)       0.3333 and size : 27x3, class : ndSparse
PY = All zero sparse: 27-by-3-by-30 and size : 27x3x30, class : ndSparse

I'm trying to do : PY(:,:,1) = A and I've got the following error :

??? Error using ==> subsasgn Attempt to convert to unimplemented sparse type Error in ==> ndSparse>ndSparse.subsasgn at 671 data=builtin('subsasgn',data,E,rhs);

Maybe can you help me to figure out why this is happening.

[Matt J]

Hmmm. Can you show me exactly what you're doing to create A and PY? And can you run the following code? When I do, I get no errors.

 A=ndSparse.build([27,3]);
 A(1,1)=1/3;

 PY=ndSparse.build([27,3,30]);
 PY(:,:,1)=A,

[Collet Bastien]

So :

PY is created by using ndSparse.build([27,3,30]);

A is created by doing :

E .* B .* C

where :

E : size = 27x3, class = double

B : size = 27x3, class = double

C : size = 27x3, class = ndSparse

[Matt J]

Can you provide a short piece of code that reproduces the error (and the exact error message that you see)? Below is yet another code example that I used to imitate what you describe. I get no error when I run it,

 E=rand(27,3);
 B=rand(27,3);
 C=ndSparse.sprand([27,3],.1);

 A=E.*B.*C;

 PY=ndSparse.build([27,3,30]);
 PY(:,:,1)=A,

[Collet Bastien]

Here is the error :

??? Error using ==> subsasgn Attempt to convert to unimplemented sparse type

Error in ==> ndSparse>ndSparse.subsasgn at 671 data=builtin('subsasgn',data,E,rhs);

Error in ==> nBayes_ETU at 175 PY(:,:,i)= A; % A = 27x3 ndSparse

Maybe can I try to print the values in your function.

[Collet Bastien]

Here are the values when the execution fails :

data =

   All zero sparse: 81-by-30

E =

    type: '()'
    subs: {':'  [1]}

rhs =

   (1,1)       0.3333

[Matt J]

I need a program that I can run myself (ideally a short one), that will let me reproduce this error.

[Collet Bastien]

I sent you a pm with all the details.

[Matt J]

I'm afraid I did not receive it.

[Collet Bastien]

Ok, can you send me an e-mail (by pm) address that I can use to send you the data.

[Matt J]

You can use the contact link here

http://www.mathworks.com/matlabcentral/fileexchange/authors/33036

Also, tell me what MATLAB version you are using.

[Collet Bastien]

I'm using the version R2009b (I forgot to write it in the pm).

[Matt J]

I don't have anything that will extract .rar files, I'm afraid. Please make it easy for me. If you can't provide a short set of commands to reproduce the problem, please just save PY and A to a .mat file and post a .zip of that. Since they are both sparse, it should give a very small, easily downloadable .zip.

I'm starting to wonder if the problem is R2009b. Formally, only R2010b and higher are supported, though I didn't suspect R2009 would have any problems. Upgrading your MATLAB version might be the simplest solution.

[Collet Bastien]

Ok,I'm looking for a new matlab version, I'm also going to send you the variables too.

[Matt J]

I received your file, but

 PV(:,:,1)=AAA;

works fine for me. I'm betting that upgrading will fix it.