Re: Structures and tensors

[Peter Cloetens]

On Aug 2, 2006, at 7:30 PM, Jordi Gutierrez Hermoso wrote:

On 02/08/06, Peter Cloetens <address@hidden> wrote:

Then I rewrote everything with cells, where each element of the cell

array corresponds to 1 matrix. This improved the speed very considerably.

I think I'm having similar issues right now with some weather

equations. Any idea what the order of magnitude of the improvement is?

As shown in the previous posts, the difference can be very, very significant.

For example:

With a 3D array:

b=zeros(1024,1024,100);a=rand(1024);tic;for k=1:100;b(:,:,k)=a;end;toc

ans = 44.664

With cells:

b=cell(100,1);a=rand(1024);tic;for k=1:100;b{k}=a;end;toc

ans = 0.0011510

In 4 dimensions:

b=zeros(1024,1024,10,10);a=rand(1024);tic;for k=1:10,for l=1:10,b(:,:,k,l)=a;end;end;toc

ans = 46.512

With cells:

b=cell(10,10);a=rand(1024);tic;for k=1:10;for l=1:10;b{k,l}=a;end;end;toc

ans = 0.0014390

You use:

 E{k}  instead of E(:,:,k)

As long as you do not need to access your data in the third dimension this

is much more efficient and readable.

How about if I need FOUR dimensions? See, I'm modelling some weather

dynamics, and I would like to keep track of the passage of time, and

I'm not assuming the atmosphere to be thin enough to consider it

two-dimensional (though I probably should). So I end up having rank 4

tensors, one for each spacetime dimension. :-/

Any suggestion how to improve this?

Thanks,

- Jordi G. H.

Peter Cloetens

ESRF

BP 220

F-38043 Grenoble

tel  : + 33 (0)4 76 88 26 50

fax : + 33 (0)4 76 88 27 85

e-mail: address@hidden

web: http://www.esrf.fr/