Re: Behavior of scatter.m

[Rik]

> Attached are three loglog plots.  The x axis is the number of points in 
> the scatter plot and the y axis is the time it took to create the 
> scatter plot.
> 
> The file badnews.pdf is done using octave on a core i7/860 with 32 bit 
> linux and 4G of RAM.
> 
> The file goodnews.pdf is done using MATLAB on a core i5/750 with Windows 
> XP and 4G of RAM.
> 
> The file bestnews.pdf is done using octave on the i7 machine above, but 
> I used 'plot(x, y, '.')' instead of scatter.
> 
> Since the windows and linux machines are pretty different and 
> octave/MATLAB do things differently, it is hard to make reasonable 
> judgments about the absolute numbers, but it is clear that octave 
> scatter performs very badly, and is in fact useless for much over 100 
> points.  In every case I get out of memory errors above 1^6 points.  
> That, at least to me, is not a problem, but I routinely do scatter plots 
> with 10,000 to 100,000 points.
> 
> I am looking at octave scatter trying to figure out where things go 
> south so badly, but I am just don't know enough about the plotting stuff 
> to make much headway.

It would be helpful if you could supply a bit more information about your
system.  Using the gnuplot toolkit, for example, will require moving data
over pipes which might be slower than the interface that OpenGL uses.  I
don't know which toolkit you were using.

In any case, I think there may be a machine-specific reason why you are
seeing this slowness.  I tried both toolkits with 3.4.0 and 1e6 points and
the time to complete was 0.12 seconds.  This was on an Intel Core 2 Duo
with 64-bit Linux and 4GB of RAM.

--Rik