Why is MATLAB's legend function so slow, and how to optimize? - matlab

Problem
In a GUI I've written I've realized that the largest bottleneck in my code performance is creating/updating the legend.
Currently I delete and recreate the legend at each update of the GUI as the user needs to be able to adjust what is in the legend. I'm currently toggling what is in the legend by adjusting the setting of LINEHANDLE.Annotation.LegendInformation.IconDisplayStyle and updating the legend using legend('off');legend('show');
Testing
The following code snippet shows that the call of legend mostly is limited by the call to legend>make_legend and is fairly independent of the legend content.
close all
n=50; % number of plot lines
S=rand(n); % data
legs=cellstr(char(26*rand(n,10)+97)); % legend entries
profile on %start profiler
plot(S)
legend(legs{:})
profile viewer % view call stats
Question
Is there a better way of updating legend content without deleting it and therefore forcing it to re-call make_legend at recreation?
Furthermore I wonder if it is known why legend in general is so slow and has such odd behavior.
Purpose
I'm adding some information here to avoid the XY Problem.
A minimal example of what I'm trying to do is:
I'm building a GUI which plots four lines, let's call them data1, data2, linear model 1, and linear model 2. The data lines are independent in both color and content, while the linear models both have the same appearance and are connected to respective data line.
I want there to be a legend which only has three entries: data1, data2, and linear model. So far, no problem.
I also want there to be three toggle-buttons which toggle the visibility of the four lines on the axes and legend. The buttons are:
data1, which toggles the visibility of both the data1, and linear model 1 data lines.
data2, which toggles the visibility of both the data2, and linear model 2 data lines.
linear model, which toggles the visibility of the linear model 1, and linear model 2 data lines.
Attempted Solutions
My first approach was to first only pass three handles to legend and then have the button callbacks adjust the visibility property of the line objects according to above.
This creates the issue that when disabling the first data line and respective linear model the legend entry for linear model also blanks out as it is connected to that specific line object, even though the other is still visible.
My current working approach instead manually sets the DisplayNameproperty of all lines and then the button callbacks adjust each lines Annotation.LegendInformation.IconDisplayStyle property. According to the documentation the user then needs to call legend to force an update.
But this is not implemented, looking at the code of legend.m it is clear that this option only returns the current legend object without any other manipulation. Therefore I'm forced to call legend('off');legend('show'); which triggers a (slow) creation of a new legend object.
This currently works but using profile I can see that the legend creation is half my computation time and has a fairly large effect on user experience when using the GUI on a slower laptop. I've already ensured that my code runs legend('off');legend('show'); only if it really has to.
The question is if any user here is able to call the unreadable, yet accessible class methods of matlab.graphics.illustration.Legend to trigger an update of an existing object without forcing it to delete and recreate. Thereby doing what the MATHWORKS documentation claims to be implemented (although it is not) in legend.
Alternatively I'm open to finding a different way of changing which line objects the current legend is tracking efficiently.

You can try to change the properties of the legend object directly. Look at http://www.mathworks.com/help/matlab/ref/legend-properties.html for a list of properties. The legend object can be accessed by:
% Assign the output of legend to a variable. Do this only once at creation.
l = legend(<some arguments>);
% Example: change the location of the legend to 'north'
l.Location = 'north';
I think this is what you asked for, but I'm not sure about any efficiency gains.

Related

Attributing multiple Peer objects to a single item in legend

Simplified Problem
I'm plotting 10 items.
To generate my plot I am creating 10 independent Line objects in a for loop
x=1:10;y=1;10;names = num2str((1:10)');
for i = 1:10
my_plots(i) = plot(x(i),y(i),'.','Color',rgb(i,:),'MarkerSize',14);
end
legend(my_plots,names);
When I click on an item in the legend it toggles visibility of the corresponding line
(legHandle.ItemHitFcn = toggleLegendItem).
function toggleLegendItem(src,evnt)
if strcmp(evnt.Peer.Visible,'on')
evnt.Peer.Visible = 'off';
else
evnt.Peer.Visible= 'on';
end
end
This works fine
I run into issues when some of my plots are actually the same category.
In this simple example, I could combine lines 1:5 into one object in a number of different ways, and pass the combined object to the legend as a single item in my_plots.
Actual problem
All of my line objects are distributed between several independent axes/subplots. Each axis has to remain independent because I need to be able to freely rotate each subplot without disturbing the other plots.
Progress so far
I've been able to build a shared legend between axes (because same class items share the same color, I only need to link the first line for each class), but I cannot figure out how to link multiple line objects to a single legend item, so that I am able to properly all lines across axes on each legend item callback.
I have an array of lines.
{{Line Line Line}}
{{Line Line Line}}
{{Line Line Line}}
...
{{Line Line Line}}
Basically what I need to do is set multiple Peer objects to a single legend item, but I am not sure if that is possible.
If someone has another solution that allows for combing multiple line handles across axes that would be helpful too.
So looking at the legend() function, I found that it's not possible to instantiate a legend with multiple Peer objects per item. There may be a way to go back and add multiple references via some sort of "combined object" but I'm not sure if a "combined object" exists that can handle objects with different Parent handles.
A quick fix to my problem (which was attributing a callback to the same item in multiple axes) was to use the shared properties within each class and across axes to get the handles for all items within each class. In this case I already assigned color to be distinctive between categories but a more objective property like Tag could also be used.
function toggleLegendItem(src,evnt)
% Find all items in this category
obj = findobj(findall(gcf,'type','Scatter'),'CData',evnt.Peer.CData);
for oid = 1:length(obj)
if strcmp(obj(oid).Visible,'on')
obj(oid).Visible = 'off';
else
obj(oid).Visible= 'on';
end
end
end
The one nuisance that this solution leaves is that the legend is a child of only one axis. If each category is not present in all axes, then the color for those missing categories may not show up. Luckily when you toggle each item's visibility the color appears and functions as normal. I would guess some other errors like this could occur.

Communication between two or more GUIs

I've a GUI(i.e. lets call it 'First')through which i can choose to open other GUIs(let's call them 'Second' and 'Third').I want to put a 'pushbutton' on the 'First'GUI that allows me to manipulate the figure on the different axes of the 'Second' and 'Third' GUIs. So, i choose with the 'First'GUI if use either 'Second' or 'Third' GUI; once i've chosen that i start to work just with the GUI that i chose (so the Third one or Second ones). Now i want to have a pushbutton not on each GUIs (Second or Third) but only on the First one in order to manipulate the figure on the axes 1 of the Second or Third (depends on which one i've previously chosen). Furthermore this pushbutton that i want is optional and i need to refresh my axes after used that.
I've done my best to explain the situation,please if you know any solution help me out!!Thanks
Your question is similar to other examples, such as this.
To understand how to solve your problem, you need to remember that MATLAB "decides" which axes to update based on the axes handle provided by the user. If the user doesn't provide a handle, a default gca (the current axes) is used. The axes that gca points to, is the first child of type 'axes' of the figure, that is, the first entry in findobj(hFigure,'Type','axes'). You can read about setting the current axes here.
Having established that, the solution you are looking for would involve storing the axes handles somewhere, and retrieving the correct one when you are about to update a plot. A common place to store it is the "application-defined data" (appdata), accessible by setappdata and getappdata, as mentioned in the first link above and also here.
The procedure you should undergo is:
Upon creating a figure, store the axes handle in appdata by calling setappdata(0,name,val) (e.g. setappdata(0,'axTag1',handles.axTag1)) from your GUI initialization function. The value 0 for the 1st argument stores it in MATLAB's root object (you can think of it as the main MATLAB window), so that even if any of the figures is closed, the information is maintained as long as MATLAB is still open.
Whenever you want to modify an axes, just obtain the appropriate handle using value = getappdata(0,name) and use it to update the corresponding axes.

Matlab Link data for a line graph on a scatter plot

I have a line which is plotted on another scatter plot. This line changes its shape from time to time. Is there any way by which I can specify in my program that the data for drawing this line is dynamic so that the plot updates by itself when the data changes?
Now what I am doing is draw the entire figure again after each data update. The program has very large number of iterations(>5000) and I need to visualize every change. That means figure should be drawn 5000 times. This is making my program very slow. Is there any other better way of doing this?
The refreshdata function might do what you want.
To automatically update a graph when a source variable changes, use the linkdata function. MathWorks has a is a great introduction page. However, there is a short example in the documentation:
x = [1:20];
y = rand(20,3);
area(x,y)
linkdata on
Then you can change a variable and the plot automatically redraws:
y(10,:) = 0;
Automatic update.
Note: Changing the source to a different variable entirely is a different thing. If YDataSource is reassigned, then refreshdata would be needed, as pointed out by Molly. Otherwise, this will keep your plot up-to-date when the variable changes.
One caveat is described on this page:
linkdata buffers updates to data and dispatches them to plots at roughly half-second intervals. This makes data linking not suitable for smoothly animating changes in data values unless they are updated in loops that are forced to execute two times per second or less.

Does Matlab execute a callback when a plot is zoomed/resized/redrawn?

In Matlab, I would like to update the data plotted in a set of axes when the user zooms into the plot window. For example, suppose I want to plot a particular function that is defined analytically. I would like to update the plot window with additional data when the user zooms into the traces, so that they can examine the function with arbitrary resolution.
Does Matlab provide hooks to update the data when the view changes? (Or simply when it is redrawn?)
While I have yet to find one generic "redraw" callback to solve this question, I have managed to cobble together a group of four callbacks* that seem to achieve this goal in (almost?) all situations. For a given axes object ax = gca(),
1. Setup the zoom callback function as directed by #Jonas:
set(zoom(ax),'ActionPostCallback',#(x,y) myCallbackFcn(ax));
2. Setup a pan callback function:
set(pan(ax),'ActionPostCallback',#(x,y) myCallbackFcn(ax));
3. Setup a figure resize callback function:
set(getParentFigure(ax),'ResizeFcn',#(x,y) myCallbackFcn(ax));
4. Edit: this one no longer works in R2014b, but is only needed if you add, e.g., a colorbar to the figure (which changes the axis position without changing the figure size or axis zoom/pan). I've not looked for a replacement. Finally, setup an undocumented property listener for the axes position property itself. There is one important trick here: We must hold onto the handle to the handle.listener object as once it's deleted (or leaves scope), it removes the callback. The UserData property of the axes object itself is a nice place to stash it in many cases.
hax = handle(ax);
hprop = findprop(hax,'Position');
h = handle.listener(hax,hprop,'PropertyPostSet',#(x,y) myCallbackFcn(ax));
set(ax,'UserData',h);
In all these cases I've chosen to discard the default callback event arguments and instead capture the axis in question within an anonymous function. I've found this to be much more useful than trying to cope with all the different forms of arguments that propagate through these disparate callback scenarios.
*Also, with so many different callback sources flying around, I find it invaluable to have a recursion check at the beginning of myCallbackFcn to ensure that I don't end up in an infinite loop.
Yes, it does. The ZOOM mode object has the following callbacks:
ButtonDownFilter
ActionPreCallback
ActionPostCallback
The latter two are executed either just before or just after the zoom function. You could set your update function in ActionPostCallback, where you'd update the plot according to the new axes limits (the handle to the axes is passed as the second input argument to the callback).

How do I use TeX/LaTeX formatting for custom data tips in MATLAB?

I'm trying to annotate a polar plot with data tips labelled with 'R:...,Theta:...' where theta is actually the Greek symbol, rather than the word spelled out. I'm familiar with string formatting using '\theta' resulting in the symbol, but it doesn't work in this case. Is there a way to apply the LaTeX interpreter to data tips? Here's what I have so far:
f1=figure;
t=pi/4;
r=1;
polar(t,r,'.');
dcm_obj = datacursormode(f1);
set(dcm_obj,'UpdateFcn',#polarlabel)
info_struct = getCursorInfo(dcm_obj);
datacursormode on
where polarlabel is defined as follows:
function txt = polarlabel(empt,event_obj)
pos = get(event_obj,'Position');
x=pos(1);
y=pos(2);
[th,r]=cart2pol(x,y);
txt = {['R: ',num2str(r)],...
['\Theta: ',num2str(th*180/pi)]};
Update: This solution is primarily applicable to versions R2014a and older, since it appears to fail for newer versions, specifically R2014b and newer using the new handle graphics system. For newer versions using the new handle graphics system, a solution can be found here.
For some odd reason, the data cursor tool in MATLAB forcibly sets the data tip text to be displayed literally instead of with TeX/LaTeX interpreting (even if the default MATLAB settings say to do so). There also appears to be no way of directly setting text properties via the data cursor mode object properties.
However, I've figured out one workaround. If you add the following to the end of your polarlabel function, the text should display properly:
set(0,'ShowHiddenHandles','on'); % Show hidden handles
hText = findobj('Type','text','Tag','DataTipMarker'); % Find the data tip text
set(0,'ShowHiddenHandles','off'); % Hide handles again
set(hText,'Interpreter','tex'); % Change the interpreter
Explanation
Every graphics object created in the figure has to have a handle. Objects sometimes have their 'HandleVisibility' property set to 'off', so their handles won't show up in the list of child objects for their parent object, thus making them harder to find. One way around this is to set the 'ShowHiddenHandles' property of the root object to 'on'. This will then allow you to use findobj to find the handles of graphics objects with certain properties. (Note: You could also use findall and not worry about the 'ShowHiddenHandles' setting)
Turning on data cursor mode and clicking the plot creates an hggroup object, one child of which is the text object for the text that is displayed. The above code finds this text object and changes the 'Interpreter' property to 'tex' so that the theta symbol is correctly displayed.
Technically, the above code only has to be called once, not every time polarlabel is called. However, the text object doesn't exist until the first time you click on the plot to bring up the data tip (i.e. the first time polarlabel gets called), so the code has to go in the UpdateFcn for the data cursor mode object so that the first data tip displayed has the right text formatting.