Re: slow 'eval' function - maybe pointers would do?

[Guido Walter Pettinari]

Hi!

Thank you for the answers, Judd & Carlo!

> You can do what you ask by:
>
> for ii = 1:N
> names{ii}  = sprintf("ks_%d", ii);
> values{ii} = somehow_compute_value (ii);
> endfor
>
> S = cell2struct (values, names, 2)


You are right, but the point is that in my code each 'ks_i' is a  
vector of M elements. I am filling all the 'ks_i' vectors inside a  
'for i=1:M; for i=1:N' nested loop. If I were to call 'cell2struct' in  
each sub-loop, I will obtain many structs, while my idea is to store  
the data in a single struct. I want to obtain a struct 's' such that  
each field 's.ks_i' contains a Mx1 vector (where M is big).

By the way, I solved the issue by memorizing the 'ks_i' into a MxN  
vector called 'ks' inside the loop. Once outside the loop, I filled  
the struct in this way:

for j=1:N
  eval(  [ 's.k', int2str(j), ' = ks ( :, j );' ]     );
endfor

In this way I use more memory but the execution is faster.

Thank you,

Guido

P.S. Carlo, is there any reason whereby in your example you used  
sprintf instead of the string concatenation operator [ 'ks_', ii ] ?  
Is the former faster?


On Oct 29, 2009, at 19:37 , Carlo de Falco wrote:

> On 29 Oct 2009, at 18:52, Guido Walter Pettinari wrote:
>
> > Hi everybody!
> >
> > I need to assign values to a series of variables with numbered names,
> > such as
> >
> > ks_1 = ... ;
> > ks_2 = ... ;
> > ...
> > ks_N = ... ;
> >
> > where N is big and is determined at runtime. I do not want to use a
> > vector named 'ks' such that ks(i) = ks_i and so on. Therefore, I used
> > the 'eval' function in this way:
> >
> > for i=1:N
> > eval ( [ 'k', int2str(i), ' = ... ;' ] );
> > endfor
> >
> > However, using 'eval' the execution time of my loop increases ~6  
> > fold.
> > To test this behaviour, please try to run the following script:
> >
> > tic();
> > for i=1:100000   sin(2);   endfor
> > toc()
> >
> > tic();
> > for i=1:100000   eval ( 'sin(2);' );   endfor
> > toc()
> >
> > On my machine, the output is:
> >
> > Elapsed time is 0.8 seconds.
> > Elapsed time is 5 seconds.
> >
> > Could you please suggest a more performance-friendly workaround?
> > Actually, my numbered variables are fields of a structure (es.
> > struct.k_1, struct.k_2 ... ); I thought that one could solve the
> > problem by using pointers to access them, but I could not find  
> > support
> > for pointers in Octave.
> >
> > Thank you,
> >
> > Guido
>
>
> You can do what you ask by:
>
> for ii = 1:N
> names{ii}  = sprintf("ks_%d", ii);
> values{ii} = somehow_compute_value (ii);
> endfor
>
> S = cell2struct (values, names, 2)
>
> but what is the reason why you don't want to just use
>
> for ii = 1:N
> ks{ii}  = somehow_compute_value (ii);
> endfor
>
> ?
>
> c.