Code in live script run much slow

The new matlab live script functionality is really awesome, thanks matlab team.

But I observe that my code run considerably slower in live script than in the command line. Just out of curiousity, is this an expected result for a newly created function or there is some way to configure the amount of resource allocation for live script? Thanks.

[Lukas Müller]

just want to add my case to the needed update in performance in matlab livescript. The following code

x1_ini = -147143866;
x2_ini = 3391949;
v1_ini = -0.68641;
v2_ini = -30.2659;
max_dev = 0.1*6371;
earth_sun = 149600000;
orbit = 2*pi*earth_sun;
v = norm(v1_ini+v2_ini);

dt = 1;
while true
    n = 1;
    while v*dt*n < (orbit-max_dev)
          n = n+1;
    end
    if v*dt*n + v*dt > (orbit+max_dev)
        break
    end
    dt = dt+1;
end
dt_max = dt-1; %= 24

takes a couple of seconds as .m file and isn't yet finished computing after 3 hours as livescript ...

[Noam Segev]

It is a bit ungainly, but you can save the file as .m file and run the heavy sections from this file, and then continue with livescript for sections that needs the represenations.

[Matt J]

Does anyone know if the speed issue extends to Live Functions?

[per isakson]

Executions of functions are not affected as far as I can see on R2018b.

I copied the code of the comment above by Lukas Müller into an ordinary m-function and a Live Editor mlx-function. tic/toc reports the same elapse time for the two

>> live_function_test
Elapsed time is 0.145617 seconds.
>> live_function_test_LE    % Live Editor
Elapsed time is 0.143375 seconds. 

P.S. I did a similar comparison with scripts. The m-script finished in about the same time as the functions. I interupted the live-script after a minute.

[feynman feynman]

"

Code in live script run much slow

" is this talking about live script on a local pc (offline) or online? Is running offline still much slower than in m files?

Answer 1

[Philip]

This issue has been partially resolved in R2020b, R2020a, R2019b Update 5, and R2019a Update 8. Most Live scripts with loops now run as fast in the Live Editor as in the Editor or at the command line.

For example, with this change, the approximate execution times of the example from Lukas Müller's post above are:

R2018b: 200.2 s

R2020a: 0.160 s

The code was timed on a Windows 10 system with a 3.6 GHz Intel Xeon W-2133 CPU.

However, as mentioned in the replies below, the performance of rendering graphics within the Live Editor is still a known issue. Specifically, figures containing many graphics objects, such as those often created by basic graphics commands (plot, line, etc.) in tight loops, can be impacted.

For example, this code (based on the post by jdg below), takes 12 seconds to run in the R2020b Live Editor, and 4 seconds in the R2020b Command Window. The code was timed on a Windows 10 system with a 3.6 GHz Intel Xeon W-2133 CPU.

for i = 1:2700
    plot(rand, rand, 'ob');
    hold on;
end
drawnow

This issue is actively being worked on. As a workaround, code that creates figures with many graphics objects can often be rewritten to reduce the number of graphics objects being created. This change can improve the overall performance of the code. For example, vectorizing the graphics code or using a different graphics command such as “scatter” can reduce the run time of the code in the Live Editor to be almost equal to the run time in the Command Window.

Just as an example, when the code above is updated to gather the data first and only plot the data once, the code now takes 0.5 seconds to run in the R2020b Live Editor, and 0.45 seconds in the R2020b Command Window. The code was timed on a Windows 10 system with a 3.6 GHz Intel Xeon W-2133 CPU.

x = zeros(1, 2700);
y = zeros(1, 2700);
for i = 1:2700
    x(i) = rand;
    y(i) = rand;
end
scatter(x, y, [], 'ob');
drawnow

If you have any other questions, or other use cases that you would like us to better consider, please feel free to reach out to Technical Support to let us know.

[Kouichi C. Nakamura]

I'm very pleased to hear about this!

[Mike Daylamani]

This is not the case. I just downloaded 2020a and it's even slower.

[Gerhard]

I can confirm Mike's comment.

[madhan ravi]

Do you also face the same in MATLAB online @Gerhard?

[Philip]

Hello,

I am sorry to hear that you are still having an issue with the performance of running in the Live Editor. That is definitely unexpected behavior. Please feel free to reach out to technical support or contact me directly to investigate the slowdown.

Thank you,

Philip

[jdg]

Agree with Mike that things are not better. I am trying out Live Editor in R2020a on a MacBook Pro running macOS Catalina (v10.15.5). After experiencing very sluggish performance I came across this page.

One thing I noticed is that evaluating runtime in the .mlx file with tic/toc is misleading. The blue circle and pause button (both indicating the script is running) are present for much longer than the script is actually running. I had to time it manually to get an accurate estimate.

For example, the code below takes ~2 seconds in an m-file on my machine.

Running it in a .mlx file, I get tic/toc results of ~2.5 seconds. However the script is in executing mode (blue dot) for anywhere from 9 - 30 seconds.

Not looking for suggestions on how to improve the code itself. It's obviously very poorly structured.

@Philip Can you try this code on your machine? I am wondering if it's a Mac issue.

clc
tic
for A_gain = [0.5:0.1:1]
    for A_phase = linspace(30,60,5)*pi/180
        for B_gain = [0.5:0.1:1]
            for B_phase = linspace(80,120,5)*pi/180
                A = A_gain*(cos(A_gain) + j*sin(A_gain));
                plot(real(A),imag(A),'b.');
                B = B_gain*(cos(B_gain) + j*sin(B_gain));
                plot(real(B),imag(B),'r.');
                
                c = A+B;
                plot(real(c), imag(c),'o');
                hold on;
            end
        end
    end
end
toc

[Sukrit Ghorai]

The lag is so much when we plot on live editor. I think my system is decent to run matlab smoothly but when it comes to live editor , any graphics on live editor such as plotting makes it lag so much. Also output inline mode is gets painful with plots (atleast my computer cant take it :( )

Answer 2

[Gijs van Oort]

The most annoying I think, is that editing is much slower too. After pressing a key, it takes noticeable time for the character to be shown on the screen. Also walking through the code with the arrow keys and scrolling are really slow. Due to this, I'm now ending up writing my code in a normal .m script and copying it to the .mlx when I'm happy with the results.

[Gijs van Oort]

In the LiveScript, I was displaying some very long symbolic equations. I found out that this greatly reduced speed and responsiveness. It was useless anyway; 90% of the displayed equation would be typeset off the screen. After I disabled the displaying of the equations (adding a ; after the assignment), the thing became much faster (but still slower than I would wanted it to be)

[Manoj]

I agree with Gijs. Live editor is way slower even on really good spec machines even with R2023b

Answer 3

[Prannay Jain]

Yes, the code runs slower in the live script than in the command line. Currently, there is no way to configure the amount of resource allocation for the live script.

I work for MathWorks and I have forwarded this feedback to the appropriate product team.

[Xinyu Zhang]

Thanks a lot!

[Ninad Chaudhari]

1 year later, The code still runs painfully slow as compared to normal "m" file or command window! Really want to use the live script feature, but the speed just kills it !

[Jan Kappen]

Any updates on that? It's still extremely slow. Not only the execution but the input/scrolling as well. Brave new world of web apps...

[Ronan Fleming]

Yes, the latency associated with editing of live scripts is frustrating. Any workaround for this?

[Pierre-Francois]

Any news on this topic? The awful sluggishness of the live script makes this otherwise very promising and appealing tool completely useless. Really too bad !

[Jan Kappen]

I have the feeling, being behind a corporate proxy makes things much worse. When I'm at home office without proxy the editor is much quicker. Do you experience the same?

[Dom Dwyer]

has this been fixed yet? we can't work with this product currently as it is too slow even on new computers

Answer 4

[Tom Anderson]

Please improve this! It's almost 50% slower to run .mlx versus .m...

I ran a code section for a live file using "Run Section" that took 72.971023 seconds. The same code .m file using "Run Section" took 52.701320 seconds. Repeated this test multiple times and .mlx is always a loser.

That's not very good. It's almost a 50% slowdown. It's not as though there's any extra interactivity between the code, either. Just waiting for the results both times, with the GUI locked up.

[Tom Anderson]

Here's a great example I found to run much slower:

tic
for t = 1:1000000
    if ~mod(t, 10000), disp(t), end
end
toc

The live code (.mlx): Elapsed time is 487.6064 seconds.

Hit F9 to run (or run as script): Elapsed time is 0.010989 seconds.

50,000 times slower? What gives?

Answer 5

[Florian Blanchet]

Hi,

Can you remove that feature or at least make it clear that is significantly slower and will reduce editor experience ?

How can you provide a feature that is so unusable without make it explicit it is still experimental ?

Answer 6

[Ryan]

I will definitely use this feature more when resource allocation improves.

Answer 7

[Frieder Wittmann]

Has this been addressed in R2018a? Didn't see anything in the release notes.

[Will Perdikakis]

Nope. Still brutally slow edits and execution.

[Frieder Wittmann]

We're at 2019b now and despite all the advertisement for the future it is still unusable slow :-(

Answer 8

[Juan José Trujillo]

Hi,

I do not know what is the motivation for you to use the live scripts. I really find this a cool way of documenting and something which approaches a bit what you can do in Jupyter Notebooks.

In fact, I have been using successfully Jupyter Notebooks with the matlab engine in the last two years to run matlab code. I have worked that way since I have some python background and did not know live scripts until recently. Just lately I have got in touch with matlabs live scripts through a colleague and we have noticed latency which I do not experience through Jupyter Notebooks. I guess the mechanics here are really different and the Notebook approach almost resembles running bare scripts with a bit of overhead on the matlab engine.

Maybe you can give Jupyter Notebook a try and get better performance and with slightly better documentation capabilities with markdown.

[Kouichi C. Nakamura]

Correct me, if I'm wrong, but I thought MATLAB engine for Jupyter notebook is not MATLAB per se. I assumed that many of newly added MATLAB functions or class constructors/methods won't run.

I really do think Markdown is superior to LiveScript's mark up. But if the most of code won't run, I cannot use Jupyter. So I have been wishing that at some point MathWorks might adopt markdown.

Or if there is a way to run official licensed MATLAB for Jupyter, then please let me know.

[Frieder Wittmann]

I use Jupyter and like it but still prefer MATLAB due to its build in debugger, profiler and access to the workspace. Also, I like that I can run commands in the command window and combine this with results from scripts.

[Juan José Trujillo]

Hi Kouichi, sorry for answering that late. I am not really into it. But, As I understand the Matlab engine in Jupyter is just a thin wrapper which passess commands straight to a headless matlab instance. In fact any time a launch a matlab notebook I get fully new instance of Matlab working in the background. So I think you get the best of both worlds, or almost depending on the application. As Frieder says, still there are several applications that you do not have access to from the Jupyter Notebook.

A really cool application of this is for instance if you work on a remote headless system with long lived sessions. You just turn on and off your graphical interface (web browser) locally and connect to the server when you need it, but the Jupyter server and matlab instance stay alive on the remote side. I think you cannot do this kind of things with the matlab IDE, or correct me if I am wrong.

[Kouichi C. Nakamura]

Thank for the information about the backstage of Jupyter. Yeah, that makes sense!