A mysterious problem regarding NaNs, imagesc and subplots.

Hi,

I am working on a project where I need to plot an error rate as a function of two parameters as both surface and imagesc. Since I do not have results for all combinations of the parameters, my matrix of error rates contains quite a few NaNs. I need to distinguish the NaNs from the zeros, and found a neat method at the end of this thread for omitting them from the imagesc plot:

    h=imagesc(A);
    set(h,'alphadata',~isnan(A))

This worked like a charm at first, but then I needed to change the axis on the surface plot (which is in a separate subplot in the same figure) to a log scale, and suddenly the NaNs in the imagesc plot pop back up. See for yourself:

    % Initialize error matrix A with random values, and some NaNs.
    A = rand(10,10);
    A(A<1/3) = NaN;

    % Here the NaNs show up as white
    figure
    subplot(1,2,1);
    surf(A)
    subplot(1,2,2);
    h=imagesc(A);
    set(h,'alphadata',~isnan(A))

    % Here they don't
    figure
    subplot(1,2,1);
    surf(A)
    set(gca,'XScale','log');
    subplot(1,2,2);
    % set(gca,'XScale','linear'); % I tried this to no avail.
    h=imagesc(A);
    set(h,'alphadata',~isnan(A))

I have no idea why this happens! Why would the axis scale on one subplot influence the other?

Is this a bug in Matlab, or am I doing something wrong here?

Any input would be much appreciated.

Answer 1

[Doug Hull]

There are three renderers in MATLAB. The only one that handles transparency is OPENGL. Unfortunately, OPENGL does not handle log scale.

[Alec Nagel]

Thanks, Doug! That takes some of the mystery away :)

I still do not quite understand why having a log scale on one subplot should affect another subplot, though...

[Walter Roberson]

There is only one renderer per _figure_.

[Alec Nagel]

Ah. I see then. Thanks.

Answer 2

[Richard]

A workaround is to use a surface with flat shading (the default) and just view it from above. You do have to do a bit of extra work to get a similar output as imagesc gives you.

    % Initialize error matrix A with random values, and some NaNs.
    A = rand(10,10);
    A(A<1/3) = NaN;

    % Here the NaNs show up as white
    figure
    subplot(1,2,1);
    surf(A)
    subplot(1,2,2);
    surf(0.5:(size(A, 2)+0.5), 0.5:(size(A, 1)+0.5), ...
        zeros(size(A)+1), A, 'edgecolor', 'none');
    axis('tight');
    view(2);
    grid('off');
    box('on');
    set(gca, 'YDir', 'reverse');

[Alec Nagel]

Thanks! That is a good suggestion.

Answer 3

[Jordan Mertes]

I'm having a similar problem. I turn my NaN to transparent but when I do this the left and bottom axis box disappear. I can't figure out how to get them back. I have tried with the other renderers but one of them makes the NaNs blue again and the other doesn't do anything. Any thoughts on this?

Thanks