Re: Octave's plotting much slower than matlab

[David Bateman]

Thomas Ilnseher wrote:
> Hi!
>
> I did the following:
>
> f=fopen('scatter.txt', 'r');
> [v c] = fscanf(f, '%d', [2, inf]);
> fclose(f);
> scatter(v(1,:), v(2,:));
>
>
> v will contain 300k points. This takes ' a few ' seconds in Matlab.
> In Octave however, I did press control-C, and afterwards used:
>
> scatter(v(1,1:2000), v(2,1:2000));
>
> which still took longer than in matlab.
>
> Bottomline: Matlab's plotting is faster by a factor of 100 !!!
>
> Is it gnuplot that I need to blame, the octave scripts to do the
> plotting, or the octave <-> gnuplot interface ?
>
> I suspect it's some .m file, because doing:
>
> octave:11> plot(v(1,:), v(2,:), 'og;hallo;');
> octave:12> legend('off');
>
> is actually _*FASTER*_ than matlab ...
>   
Doesn't surprise me given the way that Octave is constrained to
communication with gnuplot for the scatter function. Each point is a
patch object with a single point, and so Octave sends a full gnuplot
plot command to the gnuplot process for each and every point in the
scatter plot.. I suspect that this is one function that will be vastly
accelerated by the change to a different graphics backend. However, it
doesn't appear that JHandles handles the patch with a single point or
unclosed patches, and so its no use for your case.. All I can offer is,
we're working on it..

D.



-- 
David Bateman                                address@hidden
Motorola Labs - Paris                        +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 6 72 01 06 33 (Mob) 
91193 Gif-Sur-Yvette FRANCE                  +33 1 69 35 77 01 (Fax) 

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary