MATLAB Answers

0

How do I speed up plotting of interactable scatter points?

Asked by Daniel Ko on 23 Aug 2019 at 10:47
Latest activity Commented on by Daniel Ko on 24 Aug 2019 at 16:07
Hello,
I have the following code:
for ii = 1:length(tempSpikes)
p(ii) = plot(app.Trace, tempSpikes(ii)*app.msConvert, app.xi(app.m.mainCh,tempSpikes(ii)),'k*');
end
set(p, 'ButtonDownFcn', {@pointSelected, app.Trace, p})
set(app.Trace,'UserData', []);
This plot loop plots 100s to 1000s of individual markers that can be clicked on using the callback pointselected.
pointselected is a function that just flags which markers have been selected/deselected.
This plotting process takes a long time (>10 sec), and using figure tools like zoom and pan are very slow also.
Is there a way to speed this up? I tried replacing the plot command with scatter and that was 3 times as slow.
Thanks!
Edit: The plot I have has the xaxis range from 0 to 2e5 while the yaxis ranges from -1000 to 1000 at a aspect ratio of 2:1. So using ginput and distance finding is not ideal as the distance in the xaxis is much greater per pixel causing unexpected points to be chosen if the mouse is slightly off the point.
Also ideally I would like the points to be clickable and unclickable infinitely until I press a button to act on the data. I may be zooming in and out between clicking points and clicking other buttons on the GUI. ginput locks me into point collection mode.

  0 Comments

Sign in to comment.

4 Answers

Answer by Stephen Cobeldick on 23 Aug 2019 at 13:56
Edited by Stephen Cobeldick on 23 Aug 2019 at 14:17
 Accepted Answer

Actually one single plot or line call can generate multiple line objects even if each one line only contains one point, the trick is to provide both X and Y matrices as:
  • the 1st row are the data points to be plotted,
  • the 2nd row are NaN.
This forces MATLAB to recognise each column as one point. Here is a working example:
X = rand(1,1000);
X(2,:) = NaN;
Y = rand(1,1000);
Y(2,:) = NaN;
H = plot(X,Y,'*');
set(H,'ButtonDownFcn',@(o,e)set(o,'Marker','o'))
Giving this plot (without any obvious delay when plotted):
In this example I have already clicked on four points in the top left-hand corner, where the marker was changed into circles by the callback. The callback works quickly, there is no significant delay after clicking on a point.

  1 Comment

This works fantastically! Thanks. Good technique to keep in mind.

Sign in to comment.


Answer by darova
on 23 Aug 2019 at 11:08

Creates handler for each point
x = rand(3,1);
y = rand(3,1);
h = plot([x x*0]',[y y*0]','.r');
Select points
x = rand(10,1);
y = rand(10,1);
plot(x,y,'.r')
g = ginput(3);
D = pdist2([x y],g);
[~,ix] = min(D);
hold on
plot(x(ix),y(ix),'or')
hold off

  4 Comments

Show 1 older comment
Can you please accept the answer?
Why would I accept an answer that doesn't answer the question fully?
How do you want to select points then? Without using mouse?

Sign in to comment.


Answer by Yair Altman on 23 Aug 2019 at 12:55

Try to use the vectorized version of plot to plot all the data points in a single function call rather than a loop, and also try using a simpler marker object (e.g. '+' rather than '*'):
p = plot(app.Trace, tempSpikes*app.msConvert, app.xi(app.m.mainCh,tempSpikes), 'k+');
You can also try to replace the plot function with the lower-level line function, which is faster than either plot or scatter (example).
Alternatively, you could possibly also use the scatter function, which has an undocumented speedup option, which is quite old so might not work in your case, but it's worth a try (link).
Similarly, there are multiple other ways to improve plotting performance of plot/line (link).

  1 Comment

Thanks for the links.
Vectorising the plot seems to creates one line object instead of an array of line objects one for each point so the ButtonDownFcn acts on all of the points at once. What I want is for all the individual points/markers to be separate for the callback.
Am I doing the vectorisation wrong? Or is this what I should expect? Putting NaNs between values in the input arrays did not work either.

Sign in to comment.


Answer by Yair Altman on 23 Aug 2019 at 13:36

Indeed - you get a single line object with multiple data points, this is the main reason for the speedup. In your callback you can determine which data-point was clicked by checking the click-location's proximity to the data points, something similar to the following:
hAxes = gca;
clickedPos = get(hAxes,'CurrentPoint');
xSize = diff(hAxes.XLim);
ySize = diff(hAxes.YLim);
% compute distance of clicked point from all data points
dist = hAxes.PlotBoxAspectRatio(1)^2 * ((clickedPos(1,1)-p.XData)./xSize).^2 ...
+ hAxes.PlotBoxAspectRatio(2)^2 * ((clickedPos(1,2)-p.YData)./ySize).^2;
% select the data point which is relatively closest to the clicked point
[minDist, dataPointIdx] = min(dist);

  0 Comments

Sign in to comment.