Re: fast scatter plots - advice sought

[David Bateman]

David Bateman wrote:
> Jaroslav Hajek wrote:
> > On Fri, Jan 15, 2010 at 3:06 PM, dbateman <address@hidden> wrote:
> >  
> > > Ok, so given this and the feedback from Shai both scatter and the 
> > > gnuplot
> > > backend do the wrong thing here. I'll fix this over the weekend.
> > >
> > > Thanks
> > > David
> > >
> > >     
> >
> > I didn't quite follow the information flow, but I noted that if in
> > Matlab you do scatter (x, y, s, c) where x, y, c are long column
> > vectors and s is a scalar, Matlab creates a single patch that can
> > handle the varying color, through property VertexCData or something
> > like that. Is this possible for Octave? If so, then in this case one
> > would not need to split by unique colors and sizes, but just by sizes,
> > which would probably be much faster when variable colors are used.
> >
> > OTOH, when colors are specified as a full RGB Nx3 array, Matlab (2007)
> > seems to invariably make one patch per point.
> >
> >   
> The cdata is used if MarkerFaceColor or MarkerEdgeColor are "flat". So 
> yes you should be able to group all of the patches together if CData 
> is given

Thinking about it further that won't work for the gnuplot backend as it 
is written as the line objects aren't pm3d objects. The cdata coloring 
is faked with a separate line color for each value of cdata. I'm not 
sure its even possible to do everything we want if pm3d for lines either..

You could group the patches into a set of patches closest to the nearest 
colormap value, taking into account whether the mapping is direct or 
indirect, but that should be done in the backend, as the colormap might 
change between the time the set of patches are created and they are drawn..

I'm in this part of the code at the moment, so I'll take a look and see 
if I can do something to make this faster by having scatter create a 
single "flat" patch and then separate the patches at the time they are 
drawn.

D.