Optimizing a simple code

Hi People

I trying to optimize the following code and get ride of the for loop. What the code dose, is simple. It estimates the next column in a matrix based on the previous column values times a matrix.

clc

clear

close all

dt=0.01;

t=0:dt:10;

sls=zeros(2,numel(t));

M_c=[0 1;0 0];

M_c=M_c+[1 0;0 1]*1/dt;

V_s=[0;1];

for i=1:numel(t)-1

       sls(:,i+1)=(M_c*sls(:,i)+V_s)*dt;

end

Any suggestion for a built-in function that would do the for loop part?

[Geoff Hayes]

Why do you feel the need for optimization? Is the performance that poor as the code stands now?

[John D'Errico]

After responding, I see it looks like Geoff and I think alike. Ergo, my condolences are due to Geoff. :)

[AJ1]

Yeah the whole idea is to gain speed ups!!!

Answer 1

[Roger Stafford]

For your particular problem there is a way to avoid the for-loop using the colon operator, but I doubt if it is what you had in mind as a "built-in" function. Just do this:

   n = numel(t);
   s2 = 0:dt:(n-1)*dt;
   sls = [s2.*(-dt:dt:(n-2)*dt);s2];

I tend to agree with John and Geoff that the advantage of avoiding for-loops has been greatly exaggerated among many users of matlab. It is often the very best way to accomplish a given task, and as John states, even if an advantage is gained, it is often not worth the extra programming effort to achieve.

[AJ1]

I accepted this answer but actually this dose not do what the loop did!

Answer 2

[John D'Errico]

People spend too much programmer time worrying about optimizing code that does not need it. In the end, they often get something that looks impressive, but uses as much or more time to run. Or you get something that is fine in theory, but in terms of numerical computer code does something potentially nasty.

A loop is fine there unless you are running the same code millions of times, and as long as you preallocate your sls matrix to the proper size, and you have done the latter.

On my machine, that entire code fragment took 0.0136 seconds to complete. I've spent something like a thousand times that just answering this question. If we added in the time spent by you posting it, and others reading it...

If you have something that works, USE IT. Only if it becomes a problem should you then look for bottlenecks to optimize. Remember that computer time is cheap. Programmer time is not. And, well, my time is free, worth every cent you paid for it. :)

set('soapBoxMode','down')

[AJ1]

Hey All

This is just a very simplified version of the main problem! Of course this runs very quickly on every machine. But the actual problem has much bigger dimensions and will be very slow to run!

This is a very basic method to solve linear ODEs and I would be very surprised if there is not built-in function for this!

This type of problems keep comming up again and again and I dont like it when people suggest some easy but computer time very costy solutions!

Best

[John D'Errico]

THEN YOU NEEDED TO SAY THAT!

How can we know that what you asked for is not really your problem? The mind reading toolbox is still under beta test, and has not been released to us.

The fact is, there are MANY methods in MATLAB to solve basic ODEs anyway, all far better than what you have written. So if your problem is how to solve a basic ODE in MATLAB, ASK for help about that.

You might start with

odeexamples

or

help funfun

wherein you will find many ODE solvers.

If your goal is simply to rant that MATLAB does not have a tool to explicitly solve a random problem (but you won't tell us your real problem) in a poor way, but to do it more efficiently, then I won't be of much help.

By the way, while it is possible that you MIGHT have been able to solve your problem using a tool like filter in some way, but that was not the problem you wanted to solve.

[John D'Errico]

It is unfortunate that you want something that is not available to you. If you truly need speed for code to write a basic solver in a matrix form, then write it as C code. That is your option and always will be. The fact is, what you want is simply not that commonly asked for. It is just not that useful to enough people to make it something that will be provided by MathWorks. Yes, YOU think it is useful. As I said, then YOU can always write it.

Finally, there is no reason to insult people who tell you the truth. I have not attacked you personally, nor will I.

[Roger Stafford]

AJ1, ordinarily I try to stay out of flame wars, but in this case I must speak out. I have read a great many of John D'Errico's comments and answers in both this MatlabCentral Answers Forum and the comp.soft-sys.matlab Usenet Newsgroup over a span of more than a decade, and I have also communicated with him directly a number of times. I know him to be one of the most astute contributors to these groups. Moreover he has written a number of very valuable File Exchange routines.

It is most inappropriate for you to be insulting him in the manner you have done here. You are very wrong to belittle his abilities in programming - he is an excellent programmer. You really do owe him an apology. In the absence of such an apology, I for one will not be contributing to any further queries you might make here in this forum.

[Image Analyst]

AJ1's on my black list too. Even if he does apologize for saying "Ooos I wish people without solutions just stopped answering questions! Bro you simply don't know [profanity deleted] about programming!" to John, I won't answer. And of course there was at least one other prior comment with another name calling which has since been deleted.

[Roger Stafford]

This was the original insult:

Dear John! When asking for help it is important to pose the most simple problem possible, it helps every one!. It is just unfortunate when an smart ass comes with advices rather than a solution!!!

That one needs an apology too.