Re: specifying matrices before filling with values

[A S Hodel]

On Tuesday, November 12, 2002, at 08:16  PM, Mike Miller wrote:
> Here is a very important issue for efficient computation.  The use of
> 'for', as above, can sometimes be unavoidable (in other contexts), but
> I've noticed something extremely important:  The new matrix ("anomaly"
> above) should be created in advance with the desired size.  In some 
> work I
> was doing just today, I noticed that this code...
>
> for i=1:1000, rowvector = myfunction(rand(X)); Y(i,:)=rowvector; end
>
> ...ran very slowly when Y did not exist at the start of the loop.  
> Doing
> this...
>
> Y=zeros(1000,4); for i=1:1000, rowvector = myfunction(rand(X)); 
> Y(i,:)=rowvector; end
>
> ...instead sped the thing up *massively*!  IIRC, MATLAB also works this
> way.  Having to resize the matrix seems to slow things down a *lot*.
This is because, in the first case, each time a new row is created, a 
new, larger, Y must be allocated from memory, the old Y copied into the 
new, the old Y deallocated, the new data copied into the new Y, and on 
to the next loop.  In the second case, no memory 
allocation/deallocation is required at all since the first Y is large 
enough to hold all of the data.



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------