Why do plotting functions like bar or barh take a long time to execute?

I have a 1x600 vector of doubles, where each element represent the duration of a certain engineering process. I want to get a horizontal stacked bar graph which shows the duration of each process sequentially. For that I am using the following code

BHd --> 1x600 vector
BHd = [zeros(length(BHd),1), BHd]; 
figure(3)
barh(BHd', 'stacked');

However, I noticed that it takes a long time for barh function to run in this case (58.2 s). Are there any alternative ways of plotting the stacked bar graphs, that would take less time?

The plots should have the following format.

Btw, I am running MATLAB2017b on my computer

[jonas]

You can do more or less the same ting with lines, instead of patches. See here.

https://se.mathworks.com/matlabcentral/answers/413740-how-do-i-plot-input-output-data-so-as-to-have-a-continuous-bar-in-that-range-and-not-two-simple-po

You just increase the line width, and it looks like a bar. You can also plot the edges as a continous line.

[theblueeyeswhitedragon]

It does not solve my problem, because I need each patch to have distinct colors from its neighbors. I tried using a loop for doing the plotting, where I take the sum of all previous elements and add to that number the duration of the n-th process. For instance, at the 53rd iteration:

xvals = sum(BHd(1:52)) : sum(BHd(1:52))+ BHd(53);
yvals = ones(size(xvals)); 
plot(xvals, yvals, 'Color', rand(3,1));

But this approach still takes considerable time to run.

[jonas]

You dont need a loop, plot accepts multiple inputs. In the thread I linked, each segment has its own handle, and its own properties. I can show you a bit later today if no one else does

Answer 1

[jonas]

I would guess that barh is slow because it creates 600 different patch objects, each with a different handle. Try this solution with a single surface object.

%%Method 1, SURF
n=10;
figure;
subplot(1,2,1)
x=rand(1,n);
tic
X=cumsum(x);
X=[0 X;0 X];
Y=[1.1.*ones(1,n+1);0.9.*ones(1,n+1)];
Z=rand(1,n+1);
Z=[Z;Z];
surf(X,Y,Z)
view([0 90])
set(gca,'ylim',[-2 3])
t1=toc;

%%Method 2, BARH
subplot(1,2,2)
tic
x=[nan(length(x),1)'; x]; 
barh(x, 'stacked'); 
t2=toc;

%%Display performance
t2/t1

ans =

    2.9408

so surf takes 1/3 of the time in this case. If you increase the number of elements, then the difference will increase further. At 600 elements, they differ by a factor of 500 on my machine.

You can use a different colormap and a different method for calculating the Z-values, if you need more distinct color differences.

I'm guessing you cannot increase the performance further by using line objects, as I suggested in the comments. You would have to create 600 line objects, which would hurt the performance.

[theblueeyeswhitedragon]

Great solution.Thanks