Show/hide a line in uiaxes takes up to 1.5 sec

Background: I have a large application designed in AppDesigner. It contains an UIAxes. In the UIAxes I have ~40 lines (512 points each, created with semilogx). Typically I only want to plot 1-8 of these simultaneously. Checkboxes control which of the lines should be visible. The plot x,y data is pre-computed.

Problem: I want to be able to show/hide the plots without waiting seconds for anything to happen; the problem is that it takes a ridiculously long time before a click on a checkbox reflects in a plot update. Up to 1.5 seconds!

What I have tried: I have tried two methods to manage the lines:

A) Plot everything; save the handles, use set(...,'Visible',val) from the checkbox callbacks to hide/show individual lines

B) Plot-on-demand; a plot is created (data is pre-computed so no delay there) and tagged when a checkbox changes state to true, and calling delete(findobj('Tag',tag)) when a checkbox changes state to false

The performance is equally bad for both A and B. I do not understand why changing visibility is such an incredibly slow operation. Are there any other ways to do this? Have I missed any important settings? Is the complexity of my mlapp (several tabs with components) somehow a problem?

Help would be much appreciated!

[Michael Van de Graaff]

I have noticed that proliferation of tabs can be lead to slow performance, so you might try and reduce the number of tabs. but i also have an app with like 15 tabs, nested to varous degrees (not more than 4 deep nesting), and most of those tabs have UIaxes and I don't think it takes 1.5 seconds.

If you have a legend, and you don't need that to show up as fast, then ditch the legend and render it later, legends are like the slowest thing in the universe for matlab.

also, play with drawnow. sometimes matlab waits until it's less busy with other things and draws all the graphics substancially later. Today I had a figure being generated (new popup, not in the app gui) inside a for loop and i accidently ran it with 100 iterations, and the loops all finished, and the ending statement was displayed and only THEN did all 100 figure windows pop up! i'm stunned matlab did not crash.

If i put drawnow INSIDE the loop, then the graphics are updated on each loop, but the whole process will take longer overall.

also, are you passing the UIaxes handle as appropriate? always passing the target axis is just helpful to keep things straight.

try putting the code that's troublesome in a loop to magnify the problem and use the Profiler to troubleshoot. In the matlab console use the comand

profile viewer

which will open the profiler,

then press start in the profiler (top right i think)

then run the app

then press stop in the profiler.

then then profiler should show you where it spending lots of time.

[Mattias Arlbrant]

Hi Michael, Thank you very much for your reply!

• I have a single main uiaxes and below it10 tabs. Cannot reduce the number.
• I also found legends completely unusable so I am not using any (even though people keep asking me to add them).
• I have played around with various variations of drawnow but it made no difference. I guess I could probably experiment with moving it to some ideal place in the code, although I am not sure where that would be because I basically have a callback, a function call, access to pre-calculated member variables of the mlapp followed by set(...,YData
• I am not sure what you mean by "passing the UIAxes as appropriate", maybe I have missed something? Or maybe this is irrelevant since I have only one axes? What I do is set(plotHandle,'YData',val) where plotHandle = semilogx(axesHandle,X,Y)
• The profiler is not helpful because the sluggishness seems to be all in matlab's internal code.

I think the problem is in mathworks's plot handling. It seems as if it re-renders everything, including everything that is set to "invisible", after even a small change to any of the plotted lines. But even that would not explain the truly extreme delay in the plot updates.

By the way, my pc cannot be blamed, every other application runs just fine with great performace. Only matlab has problems, and only the mlapp.

[Michael Van de Graaff]

you've probably already found these but if not maybe there's somehting there:

https://www.mathworks.com/matlabcentral/answers/252979-increase-the-plotting-performance-in-the-matlab-level-drawmode-optimizing-rendering-down-sampling

https://www.mathworks.com/matlabcentral/answers/542804-how-to-dramatically-speed-up-the-function-plot

what if you just start over every time? for example, I would try something like this to start (where the data are packaged in the 1x40 struct array data)

nlines = 40;
ax = app.UIaxes;
cla(ax);
hold on
for ii = 1:nlines
    if data(ii).plot_flag
        xdata = log(data(ii).x(:)); % just to rule out semilogx being the issue
        ydata = data(ii).y(:);
        plot(ax,xdata,ydata); % obviously your data might be packaged differently
    end
end

Answer 1

[Mattias Arlbrant]

Well I tried method "B" (deleteing and re-plotting one line) above, and it was equally slow as "A". I don't think clearing and re-plotting every single line will speed things up.

[Michael Van de Graaff]

I'm not saying clear and replot the line, i'm saying clear and replot the whole figure, and only plotting the lines you need.

maybe that sounds absurd, but intuition is a tricky thing. for example, elementwise exponentian in matlab is sometimes faster in a loop

mat = rand(1000);
tic    
    mat2 = mat.^10;
toc
mat3 = ones(1000);
tic
for ii = 1:10
    mat3 = mat3.*mat;
end
toc
sum(sum(mat2-mat3)) %to show mat2 and mat3 are equal

which is not at all what I would have expected