Re: compatibility of plotting with sparse matrices

[Ben Abbott]

On Dec 13, 2012, at 3:13 PM, John W. Eaton wrote:

> Can someone confirm that Matlab returns false for the issparse call
> below?
> 
>  x = sprand (2000, 1, 0.1);
>  y = 2*x;
>  h = line (x, y);
>  issparse (get (h, 'xdata))

x = sprand (2000, 1, 0.1);
 y = 2*x;
 h = line (x, y);
 issparse (get (h, 'xdata'))

ans =

     0

> Also, what happens for setting sparse xdata, ydata, and cdata elements
> for images?  For example, what is the result of
> 
>  x = sprand (200, 200, 0.1);
>  h = image (x);
>  set (h, 'cdata', 2*x);
>  issparse (get (h, 'cdata'))

x = sprand (200, 200, 0.1);
 h = image (x);
 set (h, 'cdata', 2*x);
 issparse (get (h, 'cdata'))

ans =

     0

> If these properties are stored as full matrices, should we copy this
> behavior?

They are full.

> If we do, then plotting of sparse matrices will just "work" the same
> as plotting full matrices does now.  But we will potentially waste a
> lot of memory.
> 
> If we instead attempt to save memory by preserving sparsity of the
> property values, we have to handle sparsity in several different
> places in the plotting "backends" (so gnuplot and opengl both must be
> fixed to plot sparse arrays correctly).  And we'll introduce an
> incompatibility with Matlab, though maybe it is not important -- does
> anyone actually rely on the values returned by "get" being full
> matrices if they are set with sparse matrices?

I don't think it necessary to copy the behavior Matlab here (maybe only for 
--brain-dead).  Fixing the gnuplot backend should be fairly easy, by converting 
the properties to full matrices when needed. 

Ben