MATLAB Central Newsreader - looping/vectorising issue...

Re: looping/vectorising issue...

Roger Stafford — Thu, 12 Feb 2009 16:57:01 -0500

"Rich " <richardwood@removethishotmail.com> wrote in message <gn0vhg$rn$1@fred.mathworks.com>...
> Anyone have any suggestions on how to vectorise this loop? (by vectorise, i guess i mean converting to matrix multiplication)
>
> I'm at a loss!
>
> dimensions are:
> z1(m,n)
> z2(m,k)
> z3(m,n*k)
>
> for i=1:size(z1,1)
> z3(i,:)=reshape(z1(i,:)'*z2(i,:),1,[]);
> end
>
> thanks heaps!

[I,J] = ndgrid(1:n,1:k);
z3 = z1(:,I).*z2(:,J);

Roger Stafford

Re: looping/vectorising issue...

Rich — Fri, 13 Feb 2009 10:06:02 -0500

thanks Roger, you are a legend, and have saved me 75% of my runtime!

looping/vectorising issue...

Rich — Thu, 12 Feb 2009 11:01:04 -0500

Anyone have any suggestions on how to vectorise this loop? (by vectorise, i guess i mean converting to matrix multiplication)

I'm at a loss!

dimensions are:
z1(m,n)
z2(m,k)
z3(m,n*k)

for i=1:size(z1,1)
z3(i,:)=reshape(z1(i,:)'*z2(i,:),1,[]);
end

thanks heaps!

Re: looping/vectorising issue...

Rich — Thu, 18 Jun 2009 11:40:20 -0400

So I hit memory problems as soon as I do this for big matrices (thousands by thousands), maybe I should have mentioned that before.
So, does anyone have a less memory intensive version?

I am generally working with sparse matrices, so I think I have to kick my brain into gear enough to work out how to use the find command in this context. But any pointers would be much appreciated!

thanks

Re: looping/vectorising issue...

Rune Allnor — Thu, 18 Jun 2009 13:53:09 -0400

On 18 Jun, 13:40, "Rich " <richardw...@removethishotmail.com> wrote:
> So I hit memory problems as soon as I do this for big matrices (thousands by thousands), maybe I should have mentioned that before.
> So, does anyone have a less memory intensive version?

That's the problem with 'vectorization': You gain some speed at
the expense of memory.

If you're hitting the memory wall, you have two options:

1) Buy a bigger computer.
2) Implement a MEX routine in C or C++.

Rune